AI training and inference aren’t just power hungry—they’re ravenous. A single NVIDIA H100 server can pull 10-20 kW. Compare that to standard computing at 1-2 kW. Data centres already consume a significant chunk of US electricity, and between 2024 and 2028, that share may triple from 4.4% to 12%.
This article is part of our comprehensive guide on understanding AI data centre energy consumption and sustainability challenges, where we explore how the industry is responding to unprecedented power demands.
The grid can’t keep up. Key markets like Virginia are hitting connection limits. And here’s where it gets messy: tech companies made carbon-neutral commitments before AI exploded their power requirements.
So they’re taking action on three fronts. Microsoft is restarting Three Mile Island. Google is backing Small Modular Reactor development through Kairos Power. Amazon has thrown $500M+ at multiple nuclear partnerships. All while maintaining renewable energy PPA portfolios.
Nuclear offers something renewables can’t: baseload power that runs 24/7 without carbon emissions. No batteries needed. No weather dependency. Just continuous power for data centres that can’t tolerate interruptions.
The timelines are uncertain—2027 to 2039 deployment windows with all the regulatory and construction risks nuclear projects bring. The costs aren’t fully known. But the investments are happening because the alternative is either continued fossil fuel backup or constrained AI expansion.
If you’re planning infrastructure or evaluating cloud providers, understanding these power strategies matters. They’ll influence data centre availability, costs, and where new capacity gets built.
AI workloads need power that doesn’t stop—ever. A generative AI training cluster consumes seven or eight times more energy than typical computing. You can’t just spin down GPU clusters when the wind stops blowing.
The combination of infrastructure constraints and grid stress is driving tech giants to pursue alternative energy sources that can deliver reliable power without waiting years for grid interconnection approvals.
Natural gas provides that baseload power right now. But it produces emissions, putting carbon-neutral commitments at risk.
Microsoft, Google, Amazon, and Meta have all pledged net-zero carbon emissions within the next decade. But Microsoft’s indirect emissions increased 26% from their 2020 baseline in FY24.
Solar and wind are intermittent. Data centres aren’t. Battery storage can cover hours, maybe a day. Multi-day weather events? That’s where the economics break down.
Nuclear runs continuously. No intermittency. No storage costs. No carbon emissions. That’s why soaring power consumption is delaying coal plant closures—there simply aren’t enough renewable sources.
The interconnection queue now runs about five years. If you want guaranteed power for a new data centre, waiting isn’t viable.
Nuclear provides 24/7 carbon-free energy without grid dependency. That’s the value proposition tech companies are paying billions to secure.
Traditional nuclear plants produce over 1,000 MW from custom-built facilities that take 5-10 years to construct on-site. SMRs produce between 5 and 300 megawatts per module and get built in factories.
Factory fabrication changes everything. Components get manufactured in controlled environments, shipped as standardised modules, and assembled on-site. Construction time drops to 24-36 months.
You can scale incrementally. Start with a single module. Add more as demand grows.
Safety systems are passive—they rely on gravity and convection. X-energy’s Xe-100 uses TRISO fuel that physically cannot melt even at temperatures exceeding 1,600°C.
The trade-off: smaller reactor cores are less efficient. Only a couple modular reactors have come online, despite more than 80 commercial SMR designs currently being developed.
But SMRs enable co-located, behind-the-meter generation. You don’t need grid transmission. You don’t wait years for interconnection approval. For data centres in markets where grid capacity is maxed out, that independence is worth the efficiency penalty.
Microsoft went for speed. They signed a 20-year agreement with Constellation Energy to restart Three Mile Island Unit 1, securing 837 megawatts by 2028. The reported $1.6 billion to upgrade Three Mile Island is substantially cheaper than building from scratch.
Google is pioneering SMR technology. They made history in October 2024 with the world’s first corporate SMR purchase agreement, partnering with Kairos Power to deploy 500 megawatts. First unit online by 2030, full deployment by 2035.
Then Google doubled down. They signed a 25-year agreement with NextEra Energy to restart Iowa’s Duane Arnold nuclear plant—America’s first nuclear facility resurrection. The 615 MW plant comes back online in 2029.
Amazon is hedging everything. AWS leads with 5 gigawatts of SMR capacity by 2039 through a $500 million investment in X-energy.
The strategic differences are clear. Microsoft prioritises speed with restarts. Google pioneers new technology with Kairos while hedging with Duane Arnold. Amazon diversifies across multiple SMR vendors. All three maintain substantial renewable PPA portfolios alongside nuclear—these aren’t either/or strategies.
Baseload power is continuous electrical generation that operates 24/7 without interruption. Data centres need it because 99.99%+ uptime requirements don’t accommodate variable supply.
You can’t pause a multi-week AI training job because the wind stopped.
Solar generates only during daylight. Wind varies unpredictably. Battery energy storage systems cost $115 per kWh plus installation. Batteries alone would cost over $5 billion for a five GW facility. And batteries cover hours to maybe a day—multi-day weather events still require backup.
Nuclear reactors operate continuously at consistent output for 18-24 month cycles without emissions. That’s why nuclear energy is well matched to data centre demand.
“24/7 carbon-free energy” is emerging as a gold standard. Google is pioneering this commitment—procuring clean energy every hour of every day. It’s more demanding than annual renewable credit matching.
No mothballed nuclear plant has ever been successfully restarted in US history. Microsoft’s Three Mile Island and Google’s Duane Arnold deals are breaking new ground.
Restarts deliver power faster and cheaper than new construction. Timeline is 2-4 years versus 5-10+ years for SMRs.
But restart opportunities are limited—perhaps as few as a handful. You need plants that shut down recently for economic reasons, not safety issues.
SMRs offer different advantages. You can scale incrementally—add modules as demand grows. Technology is newer with advanced safety features.
Cost comparison is complex. Restarts have lower absolute costs but are one-time opportunities. Wood Mackenzie forecasts SMR costs falling to $120 per megawatt-hour by 2030 as manufacturers achieve learning rates.
The most effective approach? Do both. Google is pursuing Duane Arnold restart for near-term capacity and Kairos SMRs for long-term scaling.
A Power Purchase Agreement is a long-term contract—typically 10-25 years—where you commit to purchasing electricity from a specific project at predetermined prices. It’s how tech companies secure dedicated capacity and achieve cost predictability.
The big four—Amazon, Microsoft, Meta, and Google—are the largest purchasers of corporate renewable energy PPAs, having contracted over 50 GW.
PPAs enable project financing. Microsoft’s Three Mile Island restart gets funded because Microsoft committed to buying all the power it produces for 20 years.
Co-location agreements combine PPAs with behind-the-meter generation. SMRs enable grid independence, allowing data centres to operate without competing with local communities or waiting years for transmission upgrades. For organisations looking to implement renewable energy solutions into their energy strategy, PPAs provide a proven framework for securing dedicated capacity.
Economic benefits: price certainty, hedge against grid power volatility, renewable energy credit generation. Risks include technology deployment delays and regulatory uncertainties.
Traditional carbon accounting uses annual matching. You purchase renewable credits equal to total consumption over a year. Net-zero on paper. But that masks reality—you’re still consuming fossil fuel power when renewables aren’t generating.
24/7 carbon-free energy requires zero-carbon electricity supply every hour of every day. It reveals the baseload gap that solar and wind cannot fill without massive storage.
Nuclear provides 24/7 generation without storage investment. It’s the only proven carbon-free baseload technology at scale. If you’re serious about hourly matching, you need baseload generation.
When your data centre is drawing power at 3 AM on a cloudy, windless night, what’s generating that electricity? Annual matching says it doesn’t matter as long as yearly totals balance. 24/7 CFE says it must be carbon-free in that moment.
Nuclear power addresses carbon emissions but introduces a different environmental constraint: water consumption. Both nuclear reactors and data centres require substantial cooling, and combined facilities compound these demands.
Large data centres can consume up to 5 million gallons per day. Hyperscale data centres alone are expected to consume between 16 billion and 33 billion gallons annually by 2028.
Add nuclear reactors and you’ve got siting constraints. You must locate near abundant water sources—rivers, lakes, coastlines. That limits where co-located nuclear data centres can be built.
Technology can reduce consumption. Closed-loop cooling systems can reduce freshwater use by up to 70%. But “reduce” isn’t “eliminate.”
The trade-off is explicit: water consumption versus carbon emissions. Nuclear-powered data centres address climate goals but face environmental constraints around water.
Microsoft’s Three Mile Island targets 2028. Google’s Duane Arnold aims for 2029. Google’s Kairos SMRs are scheduled for first one by 2030, full 500 MW by 2035. Amazon’s X-energy partnerships span the 2030s.
All these timelines face risk factors. NRC licensing delays are common. First-of-kind projects typically face schedule overruns.
The question is whether AI demand growth will exceed nuclear deployment pace. SoftBank, OpenAI, Oracle, and MGX intend to spend $500 billion in the next four years on new data centres in the US.
The 2028-2035 window is when we’ll know if these nuclear strategies deliver or if AI expansion continues relying on fossil backup. For a complete overview of how energy demands are reshaping AI infrastructure and the sustainability challenges the industry faces, see our comprehensive guide on AI data centre energy demands and sustainability context.
Industry analysts suggest SMRs could range from $5,000-$8,000 per kilowatt of capacity, making a 300 MW installation potentially $1.5-2.4 billion. Wood Mackenzie forecasts SMR costs falling to $120 per megawatt-hour by 2030 as manufacturers achieve learning rates of 5-10% per doubling of capacity.
Modern nuclear safety standards, passive safety systems in SMR designs, and regulatory oversight aim to minimise risk. SMRs incorporate passive safety features that rely on gravity and convection rather than pumps and operator intervention. Three Mile Island Unit 1 operated safely until economic shutdown in 2019 and will undergo modern safety upgrades during refurbishment.
Theoretically yes. Practically, the economics don’t work at data centre scale. Battery storage faces prohibitive costs and technical limitations for continuous multi-day coverage. A data centre running partially on solar or wind would still need enough power to run with no renewable generation available at all.
You won’t deploy your own nuclear reactor. But you’ll experience indirect effects through grid power pricing and availability. Capacity market prices increased from $28.92/megawatt-day in 2024/2025 delivery year to $269.92/MW-day in 2025/2026. That affects what you pay for colocation or cloud services. Understanding these infrastructure bottlenecks helps you plan around cost increases and capacity constraints.
Grid constraints are already forcing innovation. Co-located generation—nuclear and renewables—avoids grid dependency. Behind-the-meter solutions eliminate interconnection queue waits. Some planned data centres will face delays or relocation. Saturated markets like Virginia are hitting limits.
They are. The big four have established substantial renewable PPA portfolios. But there simply aren’t enough renewable sources to serve both hyperscalers and existing users. Nuclear addresses what renewables plus storage cannot economically solve at scale—true 24/7 carbon-free energy.
Both are pre-commercial with different technical approaches. Neither has commercial operating reactors yet. Kairos uses molten fluoride salt cooling—a novel design requiring more development. X-energy’s Xe-100 uses pebble-bed reactors—a more established concept with Chinese precedent. X-energy has moved further through NRC design review process.
The restart involves Unit 1, which operated safely until economic shutdown in 2019, not Unit 2 which experienced the 1979 partial meltdown. Unit 1 has different containment and safety systems with decades of safe operation history.
State utility commissions regulate behind-the-meter generation and grid interconnection. Environmental agencies oversee water use and discharge permits. Nuclear-friendly states—Pennsylvania, Tennessee, Washington—are attracting investments.
Nuclear provides incremental capacity but faces deployment pace limits. Restarts are one-time opportunities—perhaps as few as a handful. SMRs require years per deployment even with modular approach. If AI demand growth significantly exceeds projections, nuclear alone cannot scale fast enough.
US nuclear technology faces export controls that could limit international deployment. Chinese SMR programmes advance without such restrictions. China’s Linglong One became the world’s first operational commercial land-based SMR in 2023. US hyperscalers drive domestic nuclear innovation, but international AI infrastructure may rely on Chinese or Russian SMR providers.
PPAs establish fixed prices for 10-25 year terms, hedging against grid electricity price fluctuations. Microsoft signed a 20-year deal for Three Mile Island power. Google signed a 25-year agreement for Duane Arnold. Those contracts provide financial predictability for long-term infrastructure planning.
The Hidden Environmental Footprint of AI Including Water Consumption and Carbon EmissionsYou probably know AI infrastructure eats electricity. Data centres consumed 4.4% of U.S. electricity in 2023, heading to 9% by 2030. But here’s what you’re missing—water consumption.
A typical AI-focused data centre burns through 300,000 gallons of water daily. That’s the hidden cost. And it’s projected to jump 870% by 2030.
This article is part of our comprehensive guide to understanding AI data centre energy consumption and sustainability challenges, where we explore the full environmental footprint beyond just electricity usage.
So you need frameworks to measure the total footprint—energy, water, and carbon. The difference between training and inference matters for your resource planning. And your cooling technology choices create trade-offs between water and energy use.
This article gives you practical measurement approaches using PUE and WUE metrics, shows you the real impacts using data from Cornell, Google, and MIT research, and walks you through optimisation strategies with 100x energy and 1000x carbon reduction potential.
AI infrastructure hits the environment in three ways: electricity, water, and carbon.
By 2030, AI data centres will emit 24 to 44 million metric tons of carbon dioxide annually—that’s like adding 5 to 10 million cars to U.S. roadways. Water usage? Just as bad: 731 to 1,125 million cubic metres per year. That’s what 6 to 10 million Americans use at home annually.
Why does AI eat so much? GPU infrastructure throws off way more heat per rack than traditional computing—30-100 kW versus 5-15 kW. All that concentrated heat needs serious cooling.
The carbon footprint breaks down into operational emissions (Scope 2—electricity you buy), facility emissions (Scope 1—backup generators and refrigerants), and supply chain emissions (Scope 3—chip manufacturing, building construction, transport). And here’s a kicker: each kilowatt hour of energy a data centre consumes requires two litres of water for cooling.
But here’s what makes measurement tricky: production systems require 15-30% idle capacity sitting around ready for spikes and failover. That overhead burns energy. You can’t just measure active computation.
The typical number is 300,000 gallons (1,135 cubic metres) daily for an AI-focused data centre. That’s driven by evaporative cooling systems needed to dump GPU heat.
At the individual query level? Google’s Gemini consumes approximately 0.26 mL water per median text prompt. Tiny per query. But billions of daily queries add up fast to facility-scale volumes.
The 870% growth projection between now and 2030 comes from AI adoption accelerating and GPU density increasing. More heat, more cooling, more water.
Water Usage Effectiveness (WUE) measures litres of water per kilowatt-hour of IT equipment energy. The typical ratio is approximately 2 litres per kilowatt-hour. Average WUE across data centres is 1.8 L/kWh, while best-in-class facilities get below 0.5 L/kWh.
In water-scarce regions, water consumption competes with agricultural and residential use. Geographic variation matters—desert facilities versus humid climate facilities have different water needs.
AI training is a one-time, computationally intensive hit with concentrated carbon cost. Inference is the ongoing operational cost.
Training a large language model generates 25-500 tonnes CO2e depending on model size and how long training takes. Big upfront hit.
Inference generates 0.001-0.01 gCO2e per query. Tiny. But it stacks up across billions of daily interactions.
Here’s the thing: cumulative inference emissions often exceed training costs within 6-12 months for popular models. The ongoing cost overtakes the upfront cost faster than you’d think.
There’s another wrinkle. Generative AI models have a short shelf-life driven by rising demand for new applications. Companies release new models every few weeks, so energy used to train prior versions goes to waste.
Training optimisation through efficient architectures and renewable energy timing offers 100-1000x carbon reduction potential. Selecting efficient ML model architectures such as sparse models can lead to computation reductions by approximately 5 to 10 times. For practical strategies on reducing carbon emissions, tech giants are increasingly turning to nuclear and renewable energy sources.
Most data centres use a combination of chillers and on-site cooling towers to stop chips from overheating.
Evaporative cooling through cooling towers gives you the highest efficiency but consumes water that cannot be reclaimed. The water evaporates—it’s gone for good.
Direct-to-chip liquid cooling delivers liquid coolant directly to GPUs and CPUs. Closed-loop systems cut facility water use and let you pack in higher density racks.
Immersion cooling submerges servers in specialised dielectric fluid. Near-zero water use. But immersion cooling entails higher upfront costs despite giving you significant energy savings.
Water-cooled data centres use less energy than air-cooled data centres. This creates trade-offs across all cooling approaches.
Geographic context matters. In water-stressed regions, priority should be low- to zero-water cooling systems to reduce direct use. In wetter regions with carbon-intensive grids, priority should be reducing power use to lower overall water consumption. These considerations tie directly into efficiency strategies when choosing your cooling approach.
PUE measures data centre energy efficiency as the ratio of total facility energy to IT equipment energy.
Here’s the formula: Total Facility Energy (IT equipment + cooling + lighting + overhead) ÷ IT Equipment Energy.
A perfect score of 1.0 means every watt goes directly to computing. Average PUE in 2022 was approximately 1.58, though high-efficiency facilities hit 1.2 or better. Industry-leading hyperscale data centres achieve PUE of 1.1-1.2.
Lower PUE means less energy wasted on non-computing stuff. Every 0.1 PUE improvement cuts energy costs proportionally. And PUE directly multiplies grid carbon intensity impact.
But PUE has limitations. Traditional metrics for data centre efficiency like PUE are insufficient for measuring AI workloads because they don’t account for energy efficiency at the intersection of software, hardware, and system levels.
WUE measures water efficiency as litres of water per kilowatt-hour of IT equipment energy.
The calculation: Annual Water Consumption (litres) ÷ Annual IT Equipment Energy (kWh).
Lower WUE is better water efficiency. Best-in-class facilities achieve WUE below 0.5 L/kWh. Average WUE across data centres is 1.8 L/kWh—that’s your baseline to beat.
WUE complements PUE by capturing the non-energy environmental dimension people overlook in efficiency discussions. It’s an emerging metric gaining importance as water scarcity increases. When evaluating the complete sustainability challenges of AI infrastructure, both metrics are essential for comprehensive assessment.
Geographic context matters a lot. WUE of 2.0 is acceptable in a water-abundant region but problematic in drought areas. Same number, different environmental impact.
There’s a distinction between consumption and withdrawal. Water withdrawal is total water taken from sources; water consumption is the portion not returned. Evaporative cooling consumes water permanently through evaporation. Closed-loop systems withdraw water but return most of it.
Evaporative cooling gives you the best energy efficiency (lowest PUE) but the highest water consumption (highest WUE). Dry cooling eliminates water use but increases energy use 15-25%, raising the carbon footprint.
Direct-to-chip liquid cooling cuts facility-level consumption while letting you deploy higher-density GPUs. Immersion cooling offers 45% energy reduction with near-zero water use but requires operational changes.
Here’s what each technology looks like:
Evaporative cooling: 2+ L/kWh water use, PUE 1.1-1.3, proven technology with geographic limitations.
Dry cooling: near-zero water, PUE 1.3-1.6, energy penalty, works best in cool climates.
Direct-to-chip: 0.5-1.0 L/kWh water, PUE 1.1-1.2, enables 100+ kW racks, higher complexity.
Immersion: near-zero water, PUE 1.05-1.15, 45% energy savings, operational transformation required.
Geographic location influences your optimal choice. In water-stressed regions, priority should be low- to zero-water cooling systems. In wetter regions with carbon-intensive grids, priority should be reducing power use. For actionable approaches to reducing your environmental footprint, consider both technology selection and workload optimisation strategies.
You need to calculate total facility energy (PUE), water consumption (WUE), and carbon emissions (Scope 1/2/3).
Google’s comprehensive approach covers: active computation + idle capacity + CPU/RAM overhead + data centre overhead (PUE) + water consumption (WUE). Their comprehensive methodology estimates a median Gemini text prompt uses 0.24 Wh energy, 0.03 gCO2e, 0.26 mL water.
Why the comprehensive approach matters: production systems require provisioned idle capacity ready to handle traffic spikes or failover that consumes energy you need to factor into total footprint.
Tools you can use: CodeCarbon estimates emissions during ML model training. MLCarbon is the most comprehensive framework for LLMs supporting end-to-end phases: training, inference, experimentation, storage.
Carbon accounting framework: Scope 1 (direct facility emissions—backup generators, refrigerants), Scope 2 (purchased electricity—the biggest chunk for AI), Scope 3 (supply chain emissions—chip manufacturing, facility construction, equipment transport, end-of-life disposal).
Common measurement mistakes: active-machine-only calculations, ignoring the water dimension, missing Scope 3.
Understanding the full environmental impact of AI infrastructure requires measurement across electricity, water, and carbon dimensions. For a complete overview of all sustainability challenges facing AI data centres, including grid stress and emerging energy solutions, see our comprehensive guide to AI data centre energy consumption and sustainability challenges.
Based on Google’s published Gemini metrics (0.26 mL per median text prompt), similar AI assistants likely use 0.2-0.5 mL water per query. At billions of daily queries, it adds up to facility-scale volumes.
AI models run on GPU/TPU processors generating significantly more heat per rack than traditional computing (30-100 kW versus 5-15 kW). All that concentrated heat needs substantial cooling, mostly through water-based evaporative systems consuming 2+ litres per kilowatt-hour.
AI has measurable environmental impact. But impact varies dramatically based on infrastructure efficiency (PUE 1.1 versus 1.8), renewable energy usage, cooling technology, and geographic location. Combined optimisation strategies show 100x energy and 1000x carbon reduction potential.
Individual AI queries have small per-interaction impact (0.001-0.01 gCO2e, 0.2-0.5 mL water), but cumulative effect at scale is substantial. A ChatGPT query consumes about five times more electricity than a simple web search. If you make 50 daily queries you’re generating roughly 180-365 gCO2e annually.
Key strategies: 1) Smart siting in renewable energy regions (73% carbon reduction potential), 2) Model optimisation through selecting efficient architectures (5-10x efficiency gains), 3) Workload scheduling during high renewable energy availability, 4) Infrastructure efficiency improvements (PUE reduction), 5) Renewable energy procurement through PPAs.
Cumulative scale represents the primary challenge. Individual improvements get offset by exponential growth in AI usage. Even if each kilowatt-hour gets cleaner, total emissions can rise if AI demand grows faster than the grid decarbonises. Projected 2030 impact: 24-44 million metric tons CO2 and 731-1,125 million cubic metres water annually for U.S. AI data centres alone.
Current trajectory is unsustainable without intervention. However, combined optimisation strategies show 100x energy and 1000x carbon reduction potential through efficient model architectures, renewable energy scheduling, and geographic smart siting. Sustainability requires: renewable energy transition, cooling technology innovation, model efficiency improvements, and geographic smart siting.
Leading companies use comprehensive lifecycle assessment including: operational energy (Scope 2), facility direct emissions (Scope 1), and supply chain/manufacturing (Scope 3). CodeCarbon measures training emissions while cloud dashboards track inference. Transparent reporting includes PUE, WUE, renewable energy percentage, and progress toward net-zero targets.
Water withdrawal is total water taken from sources; water consumption is the portion not returned. Evaporative cooling consumes water permanently through evaporation. Closed-loop systems withdraw water but return most of it, resulting in low consumption despite high withdrawal.
Yes, through dry cooling or immersion cooling technologies. Dry cooling uses air convection (near-zero water) but increases energy consumption 15-25%. Immersion cooling submerges servers in dielectric fluid, eliminating water cooling while cutting energy 45%. Trade-off is higher capital cost and operational complexity.
Location determines grid carbon intensity, water scarcity impact, cooling efficiency, and renewable energy access. Midwest and windbelt states deliver the best combined carbon-and-water profile. Cornell study identifies smart siting as the most important factor: 73% carbon reduction and 86% water reduction potential through optimal location selection.
Scope 1 covers direct facility emissions (backup generators, refrigerants). Scope 2 is purchased electricity—the biggest component for AI. Scope 3 includes supply chain emissions like chip manufacturing, facility construction, equipment transport, and end-of-life disposal. You need all three scopes for comprehensive accounting.
How AI Data Centres Are Stressing Power Grids and Creating Infrastructure BottlenecksYou’re probably thinking about deploying AI infrastructure. Maybe you’ve got plans for a new data centre, or you’re expanding an existing facility to handle GPU clusters. Either way, you need to understand what’s happening with power grids right now.
This guide is part of our comprehensive resource on understanding AI data centre energy consumption and sustainability challenges, where we explore the full spectrum of energy demands facing AI infrastructure today.
AI data centres with GPU clusters are consuming 10-20 times more electricity than traditional data centres. In Northern Virginia alone, data centres now eat up 26% of regional electricity. That’s grid stress utilities didn’t see coming and aren’t prepared to handle.
The financial impact is real. PJM capacity market prices jumped 10 times—yes, ten times—due to data centre demand. That translates to $16-18 monthly increases on residential electricity bills. And SK Group’s chairman put it bluntly: infrastructure bottlenecks, not technology, will limit how fast you can deploy AI.
So if you’re planning AI infrastructure, you’re facing three main challenges: securing grid capacity, managing your capacity market cost exposure, and navigating interconnection timelines that stretch for years.
Let’s break down what’s actually going on.
GPU clusters in AI data centres pull 2-4 times more watts per chip than traditional server processors. Generative AI training clusters consume seven or eight times more energy than typical computing workloads. When you’re training thousands of GPUs continuously for months, you create sustained high power draw that traditional grid infrastructure just wasn’t designed to handle.
Northern Virginia hosts 643 data centres consuming 26% of regional electricity. The concentration makes the stress worse because grid infrastructure was built for distributed load, not concentrated demand spikes from clustered facilities.
Here’s where it’s heading. US data centre electricity consumption is projected to jump from 183 TWh to 426 TWh by 2030—from 4% of total electricity consumption towards potentially 9%. Grid infrastructure built for distributed load can’t handle concentrated 50-100 MW point connections that hyperscale facilities demand. These infrastructure constraints are just one dimension of the broader AI data centre energy consumption challenges facing the industry.
The difference is stark. Hyperscale AI data centres consume 10-20 times more power than equivalently sized traditional facilities. A traditional enterprise data centre with 500-5,000 servers draws 1-5 MW total power. An AI hyperscale facility with 5,000+ GPU servers draws 50-100+ MW—that’s equivalent to a small city.
AI-optimised hyperscale data centres’ advanced GPU servers require two to four times as many watts to run compared to traditional counterparts. A single NVIDIA H100 GPU rack can draw 10-40 kW. Compare that to 3-5 kW for a traditional server rack.
A typical AI-focused hyperscaler annually consumes as much electricity as 100,000 households. AI training clusters need uninterrupted power while traditional workloads can tolerate scheduled maintenance windows. That flexibility difference matters when utilities are trying to manage peak demand.
Northern Virginia hosts 643 data centres—the largest concentration globally. Texas hosts 395 facilities. California has 319. That concentration creates regional grid stress that spreads way beyond the facilities themselves.
The concentration isn’t random. Existing fibre optic infrastructure hubs provide the connectivity AI workloads need. Available land accommodates large facilities. Proximity to major population centres keeps latency low.
Northern Virginia benefits from legacy telecom infrastructure and Internet2 backbone connectivity. Once facilities established the ecosystem, network effects strengthened it—self-reinforcing cycle.
Texas offers a favourable regulatory environment and lower electricity costs through ERCOT. California draws facilities despite higher costs because that’s where tech company headquarters are.
The consequence is measurable. In 2023, data centres consumed about 26% of total electricity supply in Virginia and significant shares in North Dakota (15%), Nebraska (12%), Iowa (11%) and Oregon (11%).
Electric transmission constraints are forcing some data centres to wait up to seven years or more to secure grid connections in Virginia. But the concentration continues because first-mover advantage is hard to overcome.
PJM Interconnection operates the wholesale electricity market for 13 states serving 65 million people. The capacity market is where generators commit to providing power availability 3-5 years in advance. When capacity gets tight, prices increase to incentivise new generation.
The 2025-26 PJM capacity auction showed exactly what happens when data centre demand outpaces generation capacity. Prices increased from $28.92/megawatt-day to $269.92/MW-day—approximately 10 times. The 2026/2027 auction hit $329.17/MW-day, a 22% increase.
The financial impact hits everyone. Residential electricity bills in Virginia, Ohio, Western Maryland are projected to increase $16-18 per month. Data centre operators face higher operating costs too. Capacity charges get passed through by utilities, creating OpEx uncertainty for multi-year infrastructure investments.
PJM’s market monitor was direct: “Data centre load growth is the primary reason for recent and expected capacity market conditions”.
SK Group chairman identified infrastructure, not technology or chips, as the primary bottleneck for AI deployment speed. That assessment lines up with what’s happening across multiple infrastructure layers.
Grid interconnection queues create multi-year wait times in constrained regions. Transmission infrastructure upgrades require 3-5 years because existing lines can’t handle concentrated high-power facilities without reinforcement.
Substation capacity is another constraint. Local distribution infrastructure was designed for distributed load, not 50-100 MW point connections.
New power plants require 5-10 year development timelines—can’t keep pace with demand growth. Regulatory approval processes add another 18-36 months.
Data centre supply has been constrained from the inability of utilities to expand transmission capacity due to permitting delays, supply chain bottlenecks, and infrastructure that is costly and time-intensive to upgrade. The grid will need about $720 billion of spending through 2030.
Demand response lets data centres reduce or shift power consumption during peak demand periods in exchange for financial incentives. Google has implemented workload scheduling to shift AI training tasks away from grid stress periods.
You’ve got options here. You can postpone non-urgent AI inference workloads, shift batch processing to off-peak hours, curtail cooling during mild weather, or use on-site energy storage to shave peaks.
AI inference workloads are more flexible than training. Individual inference requests can be delayed or shifted to off-peak periods without compromising model development.
Google signed utility agreements delivering data centre demand response by targeting machine learning workloads—shifting non-urgent compute tasks like processing YouTube video when the grid is strained.
Here’s an interesting stat: Grid operators could add 126 GW in new load with minimal grid capacity expansion if average annual load curtailed at rate of 1%—that’s equivalent to adding generation capacity without building new power plants.
Financial benefits include capacity market credit offsets, utility incentive payments, and avoided peak demand charges. Implementation requires workload orchestration software and real-time grid signal integration.
You’ve got four primary approaches: traditional grid connection, on-site generation, power purchase agreements, and nuclear power partnerships.
Microsoft partnered with Three Mile Island nuclear plant revival for dedicated data centre power. Constellation Energy is bringing the reactor online by 2028 under a 20-year deal. Amazon reached a deal to purchase power from another nuclear plant. Google ordered a fleet of small modular nuclear reactors, the first one completed by 2030. For a comprehensive overview of these nuclear and renewable solutions, see our detailed analysis of how tech giants are addressing energy constraints.
Nuclear energy is well matched to data centre demand because plants generate power reliably without interruption. At $1.6 billion to upgrade Three Mile Island, it’s substantially cheaper than building new plants.
Amazon, Microsoft, Meta, and Google have contracted over 50GW of renewable energy PPAs. Renewable PPAs let you procure wind and solar power directly, though intermittency means you’ll need grid backup or storage.
On-site generation using natural gas provides reliability but faces environmental regulatory challenges. Grid connection has the lowest upfront cost but the highest capacity market exposure. On-site generation has the highest capital cost but gives you operational independence.
The regulatory environment shapes your options. California mandates renewable energy reporting. Texas allows more flexible sourcing. Virginia faces grid capacity constraints that limit traditional connection options.
Grid interconnection timelines vary by region and facility size. In constrained markets like Northern Virginia, the process can require up to seven years or more—that includes interconnection application, grid impact studies, transmission infrastructure upgrades, and regulatory approvals. Less constrained regions might complete the process in 12-18 months.
In regions with significant data centre concentration, yes. PJM Interconnection projects residential bills in Virginia, Ohio, and Western Maryland will rise $16-18 per month due to capacity market price increases driven by data centre demand. The impact varies by region and grid operator. Areas with less data centre concentration see smaller or no increases.
Hyperscale data centres are massive facilities with 5,000+ servers owned and operated by single companies like AWS, Google, or Microsoft for their own workloads. They achieve economies of scale with more efficient cooling—7-15% of energy versus 30%+ for traditional facilities. Colocation facilities are shared infrastructure where multiple companies rent space and power.
While data centres can procure 100% renewable energy through power purchase agreements, the intermittent nature of wind and solar means you’ll need grid connection for backup power or significant battery storage investment. Amazon, Microsoft, Meta, and Google have contracted over 50GW of renewable energy PPAs. Google maintains grid connectivity despite renewable PPAs. Some operators are pursuing dedicated nuclear power (like Microsoft’s Three Mile Island partnership) for reliable carbon-free power.
On-site generation provides operational independence but faces challenges: high capital costs for power plants, environmental permitting complexity (especially for fossil fuel generation in states like California), multi-year development timelines, and ongoing fuel supply logistics. Many data centres use hybrid approaches—partial on-site generation plus grid connection for flexibility and redundancy.
US data centres consumed 17 billion gallons of water in 2023 for cooling systems, projected to increase to 16-33 billion gallons by 2028. AI facilities with high-density GPU clusters generate more heat than traditional data centres, requiring more intensive cooling. Hyperscale operators achieve better efficiency with cooling at 7-15% of energy versus 30%+ for enterprise facilities.
The interconnection queue backlog is the result of demand outpacing grid capacity. Transmission infrastructure requires 3-5 years to upgrade. Substation capacity was designed for distributed loads, not concentrated 50-100 MW facilities. Environmental reviews and regulatory approvals add 18-36 months. Generation capacity additions require 5-10 year development timelines. Northern Virginia—where data centres consume 26% of regional electricity—faces the longest queues in the US.
Grid capacity availability varies significantly by region. Northern Virginia, parts of Texas, and some California markets face severe constraints with multi-year interconnection queues. Less saturated markets include parts of the Midwest, Pacific Northwest (though water availability may constrain Oregon), and some southeastern states. Your site selection framework needs to evaluate grid capacity availability, interconnection timelines, capacity market costs, and regulatory environment as primary criteria.
AI training workloads require sustained high power for continuous computation and can’t be easily interrupted without compromising model development—low flexibility for demand response. AI inference workloads are more flexible. Individual inference requests can be delayed, queued, or shifted to off-peak periods without degrading training progress. This makes inference facilities better candidates for demand response programmes.
States are adopting varied approaches. Texas implemented legislation allowing grid operators to disconnect data centres during grid emergencies to protect residential service. California requires renewable energy usage reporting and is considering renewable energy mandates for new facilities. Virginia is addressing capacity constraints through grid infrastructure investments and capacity market reforms.
Infrastructure constraints have direct business planning implications. You need to factor multi-year grid interconnection delays into project timelines, budget for higher capacity market costs in your operating expenses, develop site selection frameworks that prioritise grid capacity availability over traditional cost factors, and evaluate alternative energy sourcing strategies that you’d previously only considered for reliability purposes. The bottleneck elevates infrastructure planning from an IT consideration to a boardroom strategic decision. For practical approaches to optimising energy consumption within these constraints, see our guide on cloud optimisation and efficiency strategies.
ROI for demand response participation varies by region and implementation approach. In PJM markets with high capacity prices, a 50 MW facility participating in demand response programmes can generate $500K-$2M annually in capacity credits and incentive payments—offsetting 10-30% of capacity market cost exposure. Implementation costs include workload orchestration software ($100K-$500K), grid signal integration, and operational complexity. Facilities with a high proportion of flexible inference workloads see better ROI than training-focused operations.
Power grid constraints are reshaping how we plan and deploy AI infrastructure. The infrastructure bottleneck warnings from industry leaders like SK Group reflect a fundamental shift—you can’t simply deploy AI capacity wherever you want anymore. Grid interconnection timelines, capacity market exposure, and regional electricity constraints now dictate project feasibility as much as technology choices.
For a complete overview of AI data centre energy challenges including water consumption, carbon footprint, emerging solutions, and practical optimisation strategies, see our comprehensive guide on understanding AI data centre energy consumption and sustainability challenges.
Understanding AI Data Centre Energy Consumption and Sustainability ChallengesAI infrastructure is transforming our electrical grid and environmental landscape rapidly. Understanding these changes matters if you’re making infrastructure decisions that balance innovation with sustainability.
This hub provides comprehensive context on AI data centre energy consumption, water usage, carbon emissions, and the solutions being pursued by industry leaders. Whether you’re evaluating cloud providers, planning AI workloads, or responding to stakeholder pressure on sustainability, this guide equips you with the framework, metrics, and strategic context needed for informed decisions.
Explore eight key questions covering the scale of energy consumption, environmental impacts, infrastructure constraints, and emerging solutions. Each section links to detailed cluster articles for deeper technical analysis:
U.S. data centres consumed 183 terawatt-hours (TWh) of electricity in 2024, representing 4.4% of total U.S. electricity consumption. The International Energy Agency projects this could grow to 426 TWh by 2030—a 133% increase driven primarily by AI workload expansion. To put this in perspective, a single large hyperscale AI data centre consumes enough electricity to power approximately 100,000 households, with the largest facilities approaching the energy demands of entire cities.
AI workloads consume 10-20 times more energy per square foot than traditional data centres due to GPU-intensive computing requirements. Training large language models like GPT-4 can require petaflop-days of computation, consuming megawatt-hours of electricity for a single training run. Most of the electricity used by data centres—about 60% on average—powers the servers that process and store digital information, with cooling systems accounting for 7% at efficient hyperscalers to over 30% at less efficient enterprise facilities.
The growth trajectory varies significantly by projection methodology. The IEA’s base case forecasts 426 TWh by 2030, while more aggressive scenarios accounting for rapid AI adoption suggest consumption could reach 9% of U.S. electricity by decade’s end.
Geographic concentration amplifies regional impacts. Northern Virginia, the world’s largest data centre market with 643 facilities, now dedicates 26% of regional electricity to data centre operations according to the Electric Power Research Institute—creating significant strain on local power infrastructure. Over the past year, residential power prices increased more than the national average in 8 of the 9 top data centre markets, showing how concentrated demand affects local economies.
Detailed regional impact analysis: How AI Data Centres Are Stressing Power Grids and Creating Infrastructure Bottlenecks
AI workloads are fundamentally more energy-intensive because they rely on graphics processing units (GPUs) that consume two to four times more power than traditional CPUs while running at near-constant 100% utilisation. Unlike traditional applications with variable compute patterns and idle periods, AI model training and inference operations maintain continuous high-intensity processing, transforming data centres into facilities requiring power densities exceeding 125 kilowatts per server rack compared to 6-10 kW for conventional enterprise workloads.
GPU architecture prioritises parallel processing over energy efficiency. Advanced servers at AI-optimised hyperscale data centres are equipped with GPUs that can perform trillions of mathematical calculations per second. Training large language models involves thousands of GPUs running continuously for months, leading to massive electricity consumption. Each training session can take weeks or months consuming enormous amounts of power.
AI workloads have two distinct energy profiles: training (one-time but extremely intensive) and inference (lower per-query but scales with usage). A single training run for a large language model might consume 1,000 megawatt-hours, while inference operations for deployed models collectively represent the growing share of operational energy consumption as AI adoption increases.
Doubling the amount of energy used by the GPU gives an approximate estimate of the entire operation’s energy demands when accounting for CPUs, fans, and other equipment. This rough estimate, while useful for planning, may underestimate modern AI facilities where cooling can represent 30-40% of total load for dense GPU clusters, or overestimate for highly efficient liquid-cooled deployments. The smallest Llama model (Llama 3.1 8B with 8 billion parameters) required about 57 joules per response, or an estimated 114 joules when accounting for cooling, other computations, and other demands. Comprehensive methodology estimates show a median Gemini text prompt uses 0.24 Wh of energy, 0.03 gCO2e, and 0.26 mL of water when accounting for all critical elements.
High-bandwidth memory and advanced interconnects required for AI accelerators introduce additional power overhead. Memory systems, networking, and thermal management collectively contribute 40-50% of total system power beyond the GPUs themselves. Only a handful of organisations such as Google, Microsoft, and Amazon can afford to train large-scale models due to the immense costs associated with hardware, electricity, cooling, and maintenance.
Deep dive into technical infrastructure: Understanding GPU energy requirements and data centre design implications
Beyond electricity, AI data centres have substantial water and carbon footprints often overlooked in sustainability discussions. A typical mid-sized data centre consumes approximately 300,000 gallons of water daily for cooling—equivalent to 1,000 households—while large AI facilities can require up to 5 million gallons daily (equivalent to a 50,000-person town). The carbon footprint extends beyond operational emissions to include embodied carbon from manufacturing GPUs and constructing facilities, which can represent 30-50% of total lifecycle emissions.
Hyperscale data centres alone are expected to consume between 16 billion and 33 billion gallons of water annually by 2028. Projections show water used for cooling may increase by 870% in the coming years as more facilities come online.
Water consumption has two components: direct onsite use for evaporative cooling systems and indirect consumption by power plants generating electricity for data centres. Berkeley Lab research indicates indirect water use from electricity generation typically accounts for 80% or more of total water footprint—roughly 1.2 gallons consumed per kilowatt-hour generated. Approximately 80% of the water (typically freshwater) withdrawn by data centres evaporates, with the remaining water discharged to municipal wastewater facilities.
Bloomberg research reveals that two-thirds of U.S. data centres built since 2022 are located in high water-stress regions, creating tension with residential water supplies during peak summer demand. Arizona data centres, for example, use nearly double their average water consumption during summer months when cooling demands peak and local water supplies are most constrained. The World Resources Institute forecasts that about one-third of data centres globally are now located in areas with high or extremely high levels of water stress.
By 2030, the current rate of AI growth would annually put 24 to 44 million metric tons of carbon dioxide into the atmosphere, the emissions equivalent of adding 5 to 10 million cars to U.S. roadways. AI expansion would drain 731 to 1,125 million cubic metres of water per year by 2030, equal to the annual household water usage of 6 to 10 million Americans. Carbon emissions from AI infrastructure include operational emissions (electricity generation), embodied carbon (GPU manufacturing, rare earth element mining), and end-of-life impacts (e-waste from 3-5 year server lifecycles).
Comprehensive environmental analysis: The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
Data centre concentration in specific regions is creating substantial strain on local power grids. AI data centres uniquely challenge grid operations because they can create large, concentrated clusters of 24/7 power demand. Grid stress was the leading challenge for data centre infrastructure development, with 79% of respondents saying that AI will increase power demand through 2035 due to widespread adoption. There’s currently a seven-year wait on some requests for connection to the grid.
Geographic clustering creates localised grid stress disproportionate to national averages. Northern Virginia’s “Data Centre Alley” in Loudoun County hosts the world’s highest concentration of data centre capacity, requiring utilities to expedite substation upgrades, transmission line reinforcement, and new interconnection points at accelerating pace and increasing cost. The concentration of data centres primarily in Loudoun County represents a complex challenge where current power constraints often arise due to transmission and distribution limitations rather than a lack of power generation capabilities.
Last summer, Northern Virginia narrowly avoided widespread blackouts when 60 data centres simultaneously switched to backup power in response to a grid equipment fault. Electric transmission constraints are creating significant delays for grid connections. In Ireland (the second largest data centre market in Europe), data centres account for about 21 percent of electricity demand, with EirGrid having a de facto moratorium on new data centres since 2022.
Grid capacity constraints are emerging as genuine bottlenecks to AI expansion. Industry leaders including SK Group have publicly warned that physical infrastructure limitations—not chip availability or capital—may become the binding constraint on AI scaling by 2026-2027. Utilities report 5-7 year lead times for major transmission upgrades, creating planning challenges for companies needing capacity within 18-24 months.
Peak demand is spiking as base load generation capacity contracts, while new generation projects are stuck in increasingly long interconnection queues, 95% of which consist of renewables and storage. The forecast ranges for AI data centre demand vary widely, complicating efforts to build new generation while insulating residential customers from rate increases.
Detailed grid impact analysis: How AI Data Centres Are Stressing Power Grids and Creating Infrastructure Bottlenecks
Data centres employ three primary cooling approaches—air cooling (traditional), liquid cooling (direct-to-chip), and immersion cooling (component submersion)—each representing different trade-offs between capital cost, energy efficiency, and water consumption. Air cooling with evaporative systems consumes the most water (45-60% of withdrawn water evaporated) but has lowest upfront cost. Liquid cooling improves energy efficiency by 50-70% and enables higher rack densities required for AI workloads, while immersion cooling eliminates water consumption entirely but carries the highest initial investment and operational complexity.
Air cooling remains the most common method, though its water-intensive evaporative approach is giving way to liquid cooling for AI workloads.
The shift from air to liquid cooling is being driven by AI workload density requirements, not just efficiency gains. When server racks exceed 40-50 kilowatts of power density, air cooling becomes physically impractical—airflow cannot remove heat fast enough without excessive fan power and facility design compromises. Most AI-specific data centres now deploy hybrid approaches: air cooling for general infrastructure, liquid cooling for GPU clusters.
Direct-to-chip liquid cooling and immersive liquid cooling are two standard server liquid cooling technologies that dissipate heat while significantly reducing water consumption. Liquid cooling systems utilise a coolant which absorbs heat more efficiently than air and is especially prevalent in high-performance computing. Water-cooled data centres consume less electricity than air-cooled data centres.
Immersion cooling involves bathing servers, chips, and other components in a specialised dielectric (non-conductive) fluid. This approach can eliminate 100% of direct water consumption while enabling rack densities exceeding 125 kilowatts—more than double what liquid cooling achieves. However, immersion cooling entails higher upfront costs than conventional direct liquid cooling but provides significant energy savings and space-optimisation benefits for data centre developers.
Closed-loop water systems represent an intermediate solution, recirculating coolant to reduce freshwater consumption by up to 70% compared to traditional evaporative cooling. Closed-loop cooling systems enable the reuse of both recycled wastewater and freshwater, allowing water supplies to be used multiple times. Google, Microsoft, and Amazon have committed to closed-loop designs for new facilities, though retrofitting existing infrastructure remains challenging and expensive.
In water-stressed regions, the priority should be low- to zero-water cooling systems to reduce direct use, while investing to add renewable energy to local grids to curb indirect water use. In wetter regions with carbon-intensive grids, priority should be given to reducing power use to lower overall water consumption, even if that means continued use of evaporative cooling with its higher onsite water consumption.
While cooling innovations address operational efficiency, they don’t solve the fundamental challenge of energy sourcing—which is where tech giants are making their most dramatic strategic shifts.
Technology comparison and selection guidance: The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
Implementation strategies: Reducing AI Infrastructure Energy Consumption Through Cloud Optimisation and Efficiency Strategies
Tech giants are pursuing three parallel energy strategies: nuclear power agreements (both existing plant restarts and Small Modular Reactor development), massive renewable energy Power Purchase Agreements, and emerging technologies like power-flexible AI factories that adjust computing workloads based on grid conditions. These investments reflect recognition that sustainable AI scaling requires dedicated clean energy sources beyond grid-supplied electricity, with nuclear providing baseload reliability that intermittent renewables alone cannot deliver.
Amazon, Microsoft, Meta, and Google are the four largest purchasers of corporate renewable energy power purchase agreements (PPAs), having contracted over 50GW, equal to the generation capacity of Sweden. As of 2024, natural gas supplied over 40% of electricity for U.S. data centres. Renewables such as wind and solar supplied about 24% of electricity at data centres, while nuclear power supplied around 20% and coal around 15%.
Nuclear power is gaining renewed attention as a reliable, low-carbon source of always-on electricity that could help balance intermittent renewables. Data centre operators are exploring the potential to site facilities near existing nuclear plants. Tech companies’ insatiable power demands, net-zero commitments by 2030-2040, and grid infrastructure limitations create a perfect storm favouring nuclear solutions.
Small Modular Reactors (SMRs) offering 50-300 megawatts of dedicated generation are attracting significant investment despite unproven commercial viability. These units promise faster deployment (36-48 months versus 10+ years for traditional nuclear), lower capital costs through factory production, and the ability to co-locate directly with data centres, eliminating transmission losses and enabling waste heat utilisation. Government support exceeding $5.5 billion in the United States alone, matched by billions in private investment, provides the capital needed to overcome initial deployment hurdles for Small Modular Reactors.
The sheer scale of electricity demand accelerated by AI infrastructure exceeds what any single generation type can provide within required timeframes, requiring bringing more energy online, optimising how we use it, and advancing innovative energy solutions. Power Purchase Agreements for renewable energy have become standard procurement mechanisms, but physical electricity delivery remains constrained by grid interconnection. Companies increasingly acknowledge that contracted renewable energy often cannot flow directly to data centres, raising questions about the accuracy of carbon-neutrality claims when facilities draw from grids still heavily reliant on fossil generation during peak periods or when renewable sources are unavailable.
Alphabet is developing demand response methods which allow reduced data centre power demand during periods of grid stress by shifting non-urgent computing tasks to an alternative time and location. The emergence of DeepSeek, a highly efficient AI model, highlights a new path forward: prioritising software and system-level optimisations to reduce energy consumption, achieving competitive performance with a fraction of the energy use. AI tools now predict cooling needs, optimise workload scheduling, and identify efficiency opportunities across infrastructure.
Complete solution landscape: How Tech Giants Are Pursuing Nuclear and Renewable Energy to Power AI Infrastructure
Organisations can evaluate their AI energy footprint by measuring three dimensions—infrastructure efficiency (Power Usage Effectiveness/PUE), workload efficiency (energy per inference or training task), and carbon intensity (emissions per kilowatt-hour based on energy sourcing). Reduction strategies span infrastructure selection (choosing cloud providers with low PUE and clean energy), workload optimisation (model efficiency techniques like quantisation and pruning, carbon-aware scheduling), and architectural decisions (edge computing for latency-sensitive inference, batch processing for non-urgent training). Effective approaches balance sustainability goals with performance requirements and cost constraints.
Infrastructure efficiency measurement begins with Power Usage Effectiveness (PUE), the ratio of total facility energy to IT equipment energy. While industry-leading hyperscale facilities achieve PUE values of 1.1-1.2 (meaning only 10-20% overhead), traditional enterprise data centres often operate at 1.6-1.8. However, PUE alone is insufficient for AI workloads—measuring energy per inference or per training epoch provides more relevant efficiency signals for algorithmic and architectural optimisation.
Traditional metrics for data centre efficiency like power usage effectiveness (PUE) are insufficient for measuring AI workloads as they do not account for energy efficiency at the intersection of the software, hardware, and system levels. New AI-specific metrics such as energy per AI task and grid-aware computing must be developed to ensure that AI data centres optimise energy consumption across all levels of operation.
Workload optimisation offers the highest leverage for most organisations. Model efficiency techniques including quantisation (reducing numerical precision), pruning (removing unnecessary parameters), and distillation (training smaller models from larger ones) can reduce energy consumption by 40-70% with minimal accuracy degradation. DeepSeek‘s recent demonstration of achieving competitive performance with a fraction of typical energy consumption illustrates the potential of software optimisation over hardware scaling.
Leveraging energy-efficient cloud computing generally improves datacentre energy efficiency compared to on-premise solutions due to custom warehouses designed for better PUE and carbon-free energy (CFE). Cloud platforms such as Google Cloud Platform (GCP) and Amazon Web Services (AWS) enable sustainability in AI workloads by offering tools to minimise carbon footprints. GCP allows users to select low-carbon regions based on metrics like CFE percentage and grid carbon intensity, with regions like Montréal and Finland achieving near 100% CFE.
AWS reduces the carbon footprint of AI workloads by optimising infrastructure, transitioning to renewable energy, and leveraging purpose-built silicon chips, achieving up to 99% carbon reductions compared to on-premises setups. Many engineers and organisations overlook the CFE or PUE metrics cloud platforms provide while choosing regions, often prioritising performance, cost, and meeting business metrics over sustainability.
Carbon-aware computing, which shifts workloads geographically or temporally to periods of cleaner grid energy, provides environmental benefits without efficiency compromises. This approach requires flexible workload scheduling—feasible for training jobs and batch inference, less practical for real-time customer-facing applications. Cloud providers increasingly offer carbon-aware scheduling tools, though adoption requires engineering integration and acceptance of potential delays. Non-urgent tasks can be scheduled during periods of renewable energy abundance or lower utility rates, reducing both costs and emissions.
AI tools like MLCarbon address the gap in preemptive evaluation of environmental impact during the ML model selection phase by allowing engineers to estimate the carbon footprint of different model architectures and hardware configurations before model development. To further reduce emissions and enhance sustainability in machine learning, a holistic approach that combines thoughtful model selection, continuous model optimisation, and the strategic use of energy-efficient cloud computing is crucial.
Decision framework and implementation guide: Reducing AI Infrastructure Energy Consumption Through Cloud Optimisation and Efficiency Strategies
Measurement methodologies: The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
You should track five critical metrics: Power Usage Effectiveness (PUE) for facility efficiency, Water Usage Effectiveness (WUE) for cooling impact, carbon intensity of electricity sources (grams CO2 per kWh), energy per AI task for workload efficiency, and total cost of ownership including energy expenses. These metrics collectively provide visibility into operational efficiency, environmental impact, and economic sustainability. Effective tracking requires establishing baselines, setting improvement targets, and integrating sustainability metrics into infrastructure selection and vendor evaluation processes alongside traditional performance and cost criteria.
Power Usage Effectiveness (PUE) is a metric used to determine data centre energy efficiency, calculated as Total Facility Energy divided by IT Equipment Energy. A PUE value closer to 1 indicates higher efficiency.
PUE measurement, while standardised by The Green Grid, requires careful interpretation. A facility in a cool climate with abundant renewable energy and 1.4 PUE may have lower environmental impact than a facility in a hot climate with 1.2 PUE powered by fossil fuels. Context matters—geographic location, energy sourcing, and workload characteristics all influence the meaningfulness of efficiency metrics. Beyond PUE, data centres also monitor other energy metrics to provide a comprehensive view of efficiency and to identify areas for improvement.
Energy per AI task represents an emerging metric more relevant than facility-level PUE for comparing model architectures and deployment approaches. Measuring kilowatt-hours per thousand inferences, per training epoch, or per user session enables apples-to-apples comparisons across different infrastructure choices. The CTO Advisor research on hidden GPU operational costs highlights that energy expenses often exceed hardware depreciation for continuously-operated AI infrastructure.
Carbon accounting methodology significantly affects reported sustainability performance. Scope 2 emissions (purchased electricity) are straightforward but incomplete. Scope 3 accounting for embodied carbon in manufacturing, construction, and end-of-life disposal typically adds 30-50% to total carbon footprint. Location-based versus market-based carbon accounting (renewable energy credits) can show dramatically different results for the same physical infrastructure and energy consumption.
Procuring and saving carbon footprint information needs to be automated so engineers can have easy access to this data to gain insights before training or inference phases. Tools like MLCarbon allow engineers to estimate the carbon footprint of different model architectures and hardware configurations before model development and resource-intensive training.
Green data centre practices include adoption of renewable energy, implementing energy-efficient equipment, and utilisation of hot aisle/cold aisle configurations to optimise airflow and improve cooling efficiency. A holistic approach combining thoughtful model selection, continuous model optimisation, and strategic use of energy-efficient cloud computing is crucial for reducing emissions and enhancing sustainability.
Metric selection and benchmarking: Reducing AI Infrastructure Energy Consumption Through Cloud Optimisation and Efficiency Strategies
Comprehensive footprint measurement: The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
How AI Data Centres Are Stressing Power Grids and Creating Infrastructure Bottlenecks
Detailed analysis of regional grid stress, capacity market impacts, and infrastructure limitations constraining AI expansion. Learn how Northern Virginia’s concentration of data centres is affecting electricity costs, why there are seven-year waits for grid connections, and what industry leaders like SK Group are warning about future bottlenecks.
Read time: 8-10 minutes | Type: Investigative analysis | Best for: Understanding physical constraints on AI infrastructure development
The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
Comprehensive examination of water consumption, cooling technologies, carbon accounting, and training versus inference environmental impacts. Discover why data centres consume 300,000 to 5 million gallons of water daily, how different cooling approaches compare, and how to measure your organisation’s complete environmental footprint.
Read time: 8-9 minutes | Type: Educational deep-dive | Best for: Understanding total environmental impact beyond electricity
How Tech Giants Are Pursuing Nuclear and Renewable Energy to Power AI Infrastructure
Strategic analysis of nuclear power agreements, renewable energy investments, Small Modular Reactors, and emerging technologies addressing AI energy demands. Explore Microsoft’s Three Mile Island restart, Google’s partnership with NextEra, and why hyperscalers have contracted over 50 gigawatts of renewable capacity.
Read time: 8-10 minutes | Type: Strategic analysis | Best for: Understanding industry solutions and energy strategy
Reducing AI Infrastructure Energy Consumption Through Cloud Optimisation and Efficiency Strategies
Actionable guidance on cloud provider selection, workload optimisation, carbon-aware computing, model efficiency, and sustainability measurement. Learn how to evaluate cloud providers on sustainability metrics, implement carbon-aware scheduling, and reduce energy consumption by 40-70% through software optimisation.
Read time: 10-12 minutes | Type: How-to guide | Best for: Making infrastructure decisions and implementing efficiency strategies
Data centres currently consume approximately 2% of global electricity, though this varies significantly by region. In the United States, the figure reached 4.4% in 2024 (183 TWh), with projections suggesting growth to 7-9% by 2030 depending on AI adoption rates. Some regions with high data centre concentration, such as Northern Virginia, already dedicate over 25% of electricity to data centre operations, illustrating that global averages mask significant local impacts.
AI and cryptocurrency mining have different energy profiles making direct comparison complex. Bitcoin mining consumed an estimated 120-150 TWh globally in 2024, while AI-related data centre consumption approached similar levels. However, a single training run of GPT-4 class models consumes what Bitcoin’s network uses in approximately 3-4 hours, illustrating AI’s concentrated intensity versus crypto’s distributed consumption. AI consumption is growing much faster (projected 133% increase by 2030) while cryptocurrency mining has stabilised. Additionally, AI workloads serve broader economic purposes beyond speculative value transfer, though this distinction reflects utility preferences rather than physical energy efficiency differences.
Renewable energy supply growth is unlikely to keep pace with AI data centre demand growth without significant grid modernisation and energy storage deployment. While hyperscalers have contracted over 50 gigawatts of renewable capacity through Power Purchase Agreements, the intermittent nature of wind and solar creates reliability challenges for 24/7 data centre operations. This reality drives the parallel pursuit of nuclear baseload power, with companies viewing renewables and nuclear as complementary rather than alternative strategies. Learn more: How Tech Giants Are Pursuing Nuclear and Renewable Energy to Power AI Infrastructure
The impact varies dramatically by region. In areas with high data centre concentration like Northern Virginia, the PJM Interconnection capacity market price increase (attributed primarily to data centres) translates to residential bill increases of $16-18 monthly. Over the past year, residential power prices increased more than the national average in 8 of the 9 top data centre markets. For organisations deploying AI infrastructure, energy costs typically represent 20-30% of total infrastructure costs when accounting for both direct consumption and cooling requirements. Detailed regional analysis: How AI Data Centres Are Stressing Power Grids and Creating Infrastructure Bottlenecks
Power Usage Effectiveness (PUE) is the ratio of total facility energy consumption to IT equipment energy consumption, providing a standardised measure of data centre efficiency. A PUE of 1.0 (theoretically perfect) would mean zero energy overhead for cooling, lighting, and auxiliary systems. Modern hyperscale facilities achieve 1.1-1.2, while typical enterprise data centres operate at 1.6-1.8. For AI infrastructure decisions, PUE indicates how much energy overhead your workloads incur—a facility with 1.5 PUE means your AI training job consumes 50% additional energy for facility operations beyond the GPU computation itself.
Edge computing can reduce total system energy consumption for inference workloads by processing data locally rather than transmitting to centralised data centres, though the effect varies by application. Latency-sensitive applications (autonomous vehicles, industrial automation) benefit most from edge deployment, where local processing eliminates network transmission energy and enables smaller, specialised models. However, edge infrastructure typically operates at lower efficiency (higher PUE) than hyperscale facilities, and distributed edge deployments complicate renewable energy sourcing. Edge computing is most effective as a complementary architecture for specific use cases rather than a wholesale replacement for centralised AI infrastructure.
Calculate your AI carbon footprint across three scopes: Scope 1 (direct emissions from owned generation, typically none for most organisations), Scope 2 (electricity consumption multiplied by grid carbon intensity), and Scope 3 (embodied emissions from hardware manufacturing and end-of-life disposal). For cloud-deployed AI, request carbon reporting from your provider, noting whether they use location-based (actual grid emissions) or market-based (accounting for renewable energy credits) methodology. For on-premises infrastructure, measure kilowatt-hours consumed by AI workloads and multiply by your electricity provider’s carbon intensity (typically 400-900 grams CO2 per kWh depending on regional generation mix). Tools like MLCarbon allow engineers to estimate the carbon footprint before model development. Comprehensive measurement guidance: The Hidden Environmental Footprint of AI Including Water Consumption and Carbon Emissions
Tech companies’ renewable energy claims require careful interpretation. Many companies claim “100% renewable energy” based on Power Purchase Agreements (PPAs) that contract for renewable generation equivalent to their consumption, but the contracted renewable electricity often cannot physically flow to their data centres due to grid constraints. This market-based carbon accounting approach differs significantly from location-based accounting that measures actual emissions from electricity consumed. While PPAs do fund renewable generation that displaces fossil fuels elsewhere on the grid, the geographic and temporal mismatch between contracted renewable generation and actual facility consumption means data centres frequently draw fossil-generated electricity when renewable sources are unavailable. While hyperscalers have contracted over 50GW of renewable capacity, evaluate provider sustainability claims by examining both location-based and market-based carbon reporting methodologies.
The AI infrastructure landscape is evolving rapidly, with energy and environmental considerations moving from nice-to-have considerations to genuine constraints on deployment strategies. Understanding the scale of energy consumption, recognising the full environmental footprint including water and carbon, and evaluating the solutions being pursued by industry leaders gives you the context needed to make informed decisions.
Smart siting, faster grid decarbonisation and operational efficiency could cut environmental impacts by approximately 73% (carbon dioxide) and 86% (water) compared with worst-case scenarios. The choices you make about cloud providers, cooling technologies, workload scheduling, and model efficiency directly affect both your environmental footprint and your operational costs. For example, if you’re evaluating AWS versus Google Cloud for AI workloads, the 30-40 percentage point difference in Carbon-Free Energy between their regions can translate to hundreds of tonnes of CO2 difference for large-scale deployments.
The cluster articles linked throughout this guide provide the detailed technical analysis, implementation strategies, and decision frameworks you need to navigate these challenges. Whether you’re concerned about grid constraints limiting your expansion plans, water usage in drought-prone regions, or simply want to reduce your energy bills while meeting sustainability commitments, the solutions exist—but they require informed decision-making.
Start with understanding the constraints your region faces, evaluate your current and projected AI workloads against sustainability metrics, and explore both infrastructure and software optimisation strategies. The technology landscape will continue to evolve, but the fundamental trade-offs between performance, cost, and environmental impact will persist.
The L3Harris Insider Threat Case – What the Peter Williams Guilty Plea Reveals About Protecting Trade SecretsPeter Williams, a 39-year-old general manager at L3Harris Trenchant, spent three years stealing eight zero-day exploits worth $35 million. He had security clearance. He oversaw the compartmentalised systems designed specifically to prevent this kind of theft. And he sold those exploits to Russian brokers.
It turns out clearances, compartmentalisation, and periodic audits weren’t enough. Williams walked off with proprietary cyber-weapons developed exclusively for the U.S. government and Five Eyes allies, pocketed $1.3 million in cryptocurrency, and nobody noticed until an internal investigation finally caught him three years later.
If you’re handling sensitive data or intellectual property, you’re facing similar risks. Your developers, engineers, and senior staff all have access to trade secrets, customer data, and the systems that run your business. The Williams case is a reminder that trusted personnel with legitimate access need monitoring just as much as your perimeter defences need hardening.
This article is part of our comprehensive guide on deep tech and defense innovation, where we explore the opportunities, risks, and strategic lessons from 2025’s defense sector developments. While defense technology creates enormous commercial opportunities, the Williams case illustrates the security imperative that comes with handling sensitive innovations.
So let’s examine what happened, how it happened, and what you can implement to detect threats before they cause damage.
Peter Williams pleaded guilty in October 2025 to two counts of theft of trade secrets. Over three years, he stole at least eight sensitive cyber-exploit components from L3Harris Trenchant, the defence contractor subsidiary where he worked as general manager.
He sold these exploits to Operation Zero, a Russian brokerage that calls itself “the only official Russian zero-day purchase platform.” Williams got about $1.3 million in cryptocurrency for materials that cost L3Harris $35 million in losses.
Williams wasn’t some junior developer who got greedy. He was an Australian national who previously worked at the Australian Signals Directorate before joining L3Harris. He had the credentials and the position to access highly sensitive materials.
From 2022 through 2025, Williams conducted his transactions via encrypted communications and bought luxury items with the proceeds. He’s looking at up to 20 years, with sentencing guidelines suggesting 87 to 108 months.
Prosecutors are seeking forfeiture of his residence, luxury watches, jewellery, and the funds sitting in seven bank and cryptocurrency accounts.
Williams exploited his general manager position to access cyber-exploit components across compartmentalised systems. His role granted privileged access to sensitive systems that would normally stay isolated from each other.
He extracted materials over three years using encrypted communications channels that bypassed standard data loss prevention systems. It took three years to detect him, which tells you L3Harris didn’t have continuous behavioural monitoring running during the exfiltration period.
Here’s the problem with compartmentalisation: it assumes people stay within their assigned boundaries. When the insider manages those compartments, your strategy collapses. And without behavioural monitoring to flag unusual access patterns, periodic audits won’t catch ongoing theft before serious damage is done.
There’s another detail that makes this worse. Williams oversaw an internal investigation into suspected leaks while conducting his own theft. His supervisory position let him avoid scrutiny—a scenario that proper separation of duties and independent oversight would prevent.
Zero-day exploits target software vulnerabilities that vendors don’t know about, making them undetectable by standard defences. Williams wasn’t taking theoretical research—he extracted working attack tools ready for operational deployment.
L3Harris Trenchant developed zero-days exclusively for U.S. government and Five Eyes allies—Australia, Canada, New Zealand, the United Kingdom, and the United States. These exploits provide offensive cyber capabilities for intelligence gathering and targeted attacks.
The Department of Justice valued the eight stolen exploits at $35 million. Williams sold the first for $240,000 and agreed to sell seven more for $4 million total, though he only received $1.3 million before getting caught.
The value comes from exclusivity. Once you use a zero-day, security researchers can identify it, vendors can patch it, and effectiveness drops to zero. Operation Zero offers $200,000 to $20 million for high-value exploits, which gives you an idea of the demand from nation-states.
Operation Zero markets itself as “the only official Russian zero-day purchase platform”. The organisation acquires exploits from security researchers and insiders, then resells them to non-NATO buyers including Russian government entities.
Williams signed multiple contracts outlining payments and support fees totalling millions in cryptocurrency. The brokerage provides plausible deniability for Russian intelligence while acquiring restricted Western capabilities.
This is state-sponsored economic espionage with a commercial façade.
Williams extracted materials over three years without triggering detection systems. That timeline reveals multiple missed opportunities to identify and investigate suspicious behaviour before he caused significant damage.
He used encrypted communications to conduct transactions with Operation Zero. When privileged users access encrypted channels that aren’t approved for work, that should trigger an investigation. Particularly when those channels enable data exfiltration that bypasses standard monitoring.
Williams oversaw an internal investigation into suspected leaks while conducting his own theft—a conflict of interest that proper separation of duties would have prevented. When the people who investigate threats are themselves the threats, your governance structure has failed.
Here’s what effective monitoring would flag:
Traditional security clearance processes assume vetted individuals remain trustworthy indefinitely. The Williams case proves that assumption wrong.
User and Entity Behavior Analytics (UEBA) platforms leverage AI to detect patterns without needing predetermined indicators. UEBA establishes what normal looks like for each employee during a 30-90 day learning period, then flags deviations without requiring predefined rules.
Data Loss Prevention (DLP) monitors data movement across email, USB, cloud, and network channels. While UEBA focuses on user behaviour, DLP focuses on data behaviour—where sensitive information goes and whether movement complies with your policies.
Effective programs integrate both approaches. UEBA establishes baselines and reduces false positives through continuous learning. DLP prevents actual exfiltration when suspicious activity begins. Human analysis provides context to distinguish legitimate business activities from actual threats.
Continuous monitoring observes user actions in real-time rather than through periodic audits. Periodic audits only catch threats after the damage is done. Continuous monitoring lets you intervene before theft is complete.
The Williams case would have triggered multiple UEBA alerts: cross-compartment access, after-hours usage, encrypted communications, and data anomalies. Any one of those might have a legitimate explanation. All of them together demand investigation.
The defense sector risks illustrated by the Williams case apply equally to commercial technology companies handling valuable intellectual property. Effective programs require formalised structure with executive sponsorship, dedicated resources, and integration across departments. Carnegie Mellon’s framework addresses 13 key elements including organisation-wide participation, oversight, confidential reporting, and incident response plans.
Start by identifying your sensitive data, establishing your risk tolerance, and documenting policies. You can’t protect what you don’t know exists.
Access controls form the foundation. Implement least privilege, role-based access, and privileged access management (PAM). Every user gets the minimum access required. When roles change, access changes. Privileged accounts require session recording and approval workflows.
Detection technologies include UEBA for behavioural analytics and DLP for data movement. Commercial UEBA costs $5-15 per user monthly, enterprise DLP ranges $20-40 per user monthly for companies with 50-500 employees.
Your policy frameworks need to cover acceptable use, monitoring transparency, incident response, and employee consent. Monitoring without transparency destroys trust. State clearly what gets monitored, why, and how investigations work.
Audit logging captures privileged activities, data access, and system modifications. Make sure logs retain long enough to detect long-term threats.
Frame programs as protective rather than punitive. If employees perceive monitoring as surveillance, they’ll resist it.
For SMBs, start with logging and basic DLP using tools you already have. Move to UEBA and PAM as your budget matures. Advanced zero trust implementations require significant investment but defend against sophisticated threats.
The Williams case teaches you this: even with compartmentalisation and security clearances, a single insider can inflict massive damage. Continuous behavioural monitoring, strict privileged access governance, and evidence-based investigations aren’t optional.
Transparency about monitoring builds trust while enabling security. State clearly what gets monitored, why, and how the organisation uses monitoring data. When there’s clear communication and demonstrated responsibility, 71% of employees trusted their employers to deploy AI ethically.
Focus monitoring on high-risk activities rather than invasive surveillance. Privileged access to sensitive systems warrants monitoring. Normal business communications do not.
Use privacy-preserving techniques: anonymised baselines, threshold-based alerting, and human review before identification. UEBA systems flag anomalous behaviour without immediately identifying users. Individual identification only happens when behaviour crosses investigation thresholds.
Over 140 countries have comprehensive privacy legislation. Your implementation needs to comply with GDPR, CCPA, and other frameworks.
Investigation protocols should establish reasonable suspicion requirements, legal review, HR collaboration, and evidence preservation. Clear protocols protect both your organisation and your employees.
The Williams case shows security clearances alone create false trust. Monitoring becomes necessary even for vetted personnel. But that monitoring needs to be transparent, proportionate, and focused on legitimate security concerns.
Communicate the “why” behind monitoring. You’re protecting company assets, customer data, and employee jobs. When competitors steal trade secrets or ransomware groups exfiltrate data, everyone loses.
Only 21% of consumers trust tech companies to protect their data. Your employees understand breaches happen and know monitoring serves protective purposes. What they won’t accept is surveillance extending into productivity tracking or personal communications.
The balance isn’t between security and trust—it’s between transparent, proportionate security that builds trust and opaque surveillance that destroys it.
The Williams case demonstrates that innovation security is just as critical as technological innovation itself. For a complete overview of how security considerations fit within the broader landscape of deep tech opportunities and strategic lessons from 2025’s defense sector, see our comprehensive deep tech and defense innovation guide.
An insider threat is when someone with authorised access uses it maliciously or negligently to cause harm. Unlike external attackers who need to breach perimeter defences, insiders already have legitimate credentials, making detection more challenging. The Williams case shows this perfectly—a trusted employee who exploited privileged access for financial gain. Most insider incidents are unintentional, but malicious cases cause disproportionate damage because insiders know where valuable assets live and understand the security controls they need to circumvent.
Williams was charged with two counts of theft of trade secrets under 18 U.S.C. § 1832, each carrying a maximum 10-year prison sentence. Federal sentencing guidelines suggest 87 to 108 months, meaning roughly 7-9 years imprisonment. He faces restitution of $1.3 million plus asset forfeiture including his residence, luxury watches, jewellery, and cryptocurrency accounts.
Start with tools you already have. Native cloud audit logging comes included with platforms you’re already paying for. Open-source DLP and basic access controls cost minimal additional investment. Intermediate implementations adding commercial UEBA ($5-15 per user monthly) and enterprise DLP ($20-40 per user monthly) will run you $15,000-50,000 annually for companies with 50-500 employees. Advanced programs with zero trust and PAM reach $75,000-150,000 annually. The Williams case’s $35 million loss shows even modest programs deliver strong ROI.
Yes, through transparency, consent, and compliance. Employers can monitor work systems if employees are informed through clear policies and provide consent. GDPR Article 25 requires appropriate technical and organisational measures during system design. The key requirements: disclose what gets monitored, focus on work-related activities not personal communications, and comply with regional privacy laws. You’ll need legal review because requirements vary by location and industry.
L3Harris relied on clearances and compartmentalisation without implementing continuous behavioural monitoring. The key failures: no UEBA system to flag unusual access patterns, insufficient audit logging of privileged activities, periodic rather than continuous monitoring (which allowed three years of undetected theft), and over-reliance on security clearances creating false trust. Williams’s supervisory position during an internal investigation he oversaw was a conflict of interest that proper separation of duties would have prevented.
UEBA focuses on behavioural anomalies, using machine learning to establish baselines and flag suspicious actions. UEBA platforms detect patterns without predetermined indicators. DLP monitors data movement—emails, uploads, USB transfers—blocking or alerting on policy violations based on content inspection. UEBA provides early warning by detecting behavioural changes before data loss happens. DLP prevents the actual theft during exfiltration. You need both working together.
Consult legal counsel immediately to ensure you comply with employment law and preserve evidence properly. Document specific suspicious behaviours without confronting the employee prematurely. Engage HR to review personnel records and behavioural changes. Preserve digital evidence through forensic copies of systems and audit logs. Legal counsel must review decisions to ensure privacy compliance. Consider temporary access restrictions if theft is ongoing, balancing security with legal risks. Only after legal and HR review should you move to confrontation or termination.
A starter program—audit logging, basic DLP, access control review—launches in 4-8 weeks: 1-2 weeks for policy and legal review, 2-3 weeks for deployment, 1-2 weeks for training. Intermediate programs adding UEBA and PAM need 3-6 months. UEBA requires 30-90 days to establish baselines, while access restructuring introduces complexity. Advanced programs with zero trust span 6-12 months and involve architectural changes. Start with quick wins while you plan longer-term capabilities.
Statistically, insider threats cause greater average damage. Verizon’s Data Breach Investigations Report shows insiders are involved in 20-30% of breaches but cause disproportionate impact. Insiders have legitimate access, know where assets live, understand the controls they need to circumvent, and stay undetected longer. Williams operated for three years before detection. External attacks happen more frequently overall. Your optimal security strategy addresses both: perimeter defences for external threats, behavioural monitoring for insiders.
Zero trust assumes no user is inherently trusted. Every access request gets verified based on identity, device health, context, and least privilege. Unlike perimeter security, zero trust continuously validates through multi-factor authentication, micro-segmentation limiting lateral movement, real-time risk assessment, and comprehensive logging. This restricts access even for authenticated users. Williams couldn’t have accessed all those compartments under a zero trust model. However, implementation requires significant architectural changes, making it a longer-term goal for most SMBs.
Leverage cloud-native tools. Microsoft 365 and Google Workspace offer native DLP and audit logging. Cloud access security brokers monitor SaaS usage. Endpoint detection tracks device activities. Managed security providers offer outsourced monitoring at $2,000-5,000 monthly, which is cheaper than hiring full-time staff. Effective SOCs can be built using automation to reduce workload. Prioritise high-impact controls: strict access management, mandatory multi-factor authentication, automated audit logging, and basic DLP. The goal is risk reduction, not perfection.
Core technologies include UEBA platforms (Exabeam, Securonix, Microsoft Sentinel) for detecting behavioural anomalies. DLP systems (Forcepoint, Symantec, Microsoft Purview) monitor data movement. Privileged access management tools (CyberArk, BeyondTrust) record admin activities. Endpoint detection tools (CrowdStrike, SentinelOne) track file access. SIEM platforms (Splunk, Elastic) aggregate logs for investigation. Next-generation data detection leverages data lineage to understand how user actions impact sensitive information. These technologies work together: UEBA flags unusual patterns, DLP blocks unauthorised transfers, and PAM records privileged activities for forensics.
Defense Tech Investment in 2025 – Where Government and Venture Capital Are Backing Breakthrough InnovationDefence tech investment hit $38 billion through the first half of 2025 while overall VC funding declined. That’s market resilience worth paying attention to.
This analysis is part of our comprehensive guide on deep tech and defense innovation, examining the opportunities, risks, and strategic lessons shaping the sector in 2025.
Government co-investment programs have changed the game. The Office of Strategic Capital, SBIR/STTR, and STRATFI/TACFI are creating hybrid funding models that bridge public and private capital. Investment is flowing into autonomous systems, AI/ML, cybersecurity, and hypersonics.
For you, this landscape reveals strategic partnership opportunities and dual-use technology potential. Understanding how these funding mechanisms work helps you make better build vs buy decisions.
The major players? Established VC firms like Founders Fund and Sands Capital, startups like Anduril at $14 billion valuation, and new government funding mechanisms that are changing the game.
Global security tensions. Ukraine. Indo-Pacific competition. Governments are funding solutions and treating defence modernisation as a top priority.
DoD‘s FY2026 budget proposes $832 billion in discretionary funding, including $148 billion for research, development, test and evaluation. That’s real money flowing to emerging technologies and non-traditional contractors.
Dual-use technology is the key. It lets defence tech companies serve both government and commercial markets. Defence contracts provide the stability, commercial sales drive the volume and scale.
Government co-investment programs take the edge off investor risk. The Office of Strategic Capital, STRATFI, and TACFI match private capital with public funding. Anduril’s $1.5 billion Series F demonstrates there’s a viable path to scale and profitability.
Traditional prime contractors are partnering with or acquiring startups. That creates exit opportunities for investors. And commercial technology adaptation? It’s proving faster and more cost-effective than the traditional defence R&D cycles.
Defence tech funding reached $3 billion in 102 deals in 2024, an 11% uptick from 2023. The investor base grew from fewer than 100 firms in 2017 to more than 300 different firms in 2024. We’re talking Andreessen Horowitz, Alumni Ventures, 8VC, Founders Fund, and General Catalyst.
OSC provides loans and loan guarantees. Not equity investments. This helps defence tech companies attract private capital without diluting ownership.
Here’s the typical structure – your company raises a private funding round, then OSC provides a matching loan to reduce dilution and extend your runway. Direct loans are available up to $150 million to finance projects in the United States.
To be eligible you need to be developing technologies aligned with the National Defense Science and Technology Strategy. You’ll need to demonstrate private investor interest and show a pathway to DoD procurement.
Loan terms include below-market interest rates and flexible repayment tied to contract milestones. Convertible structures are available too. The advantage? Accessing growth capital without surrendering equity or control to government entities.
OSC has deployed over $1 billion since 2022 across autonomous systems, AI, and advanced manufacturing.
STRATFI and TACFI work a bit differently. They provide DoD matching funds to private investment. STRATFI handles the larger awards – $5M to $50M – for strategic technologies and multi-year programs. TACFI provides smaller awards, $1M to $10M, for tactical solutions with faster deployment timelines.
Both require a 1:1 private capital match, DoD end-user validation, and a pathway to Program of Record. Applications go through service innovation hubs. AFWERX handles Air Force STRATFI. NavalX handles Navy programs.
SBIR Phase I gets you $50,000 to $275,000 over 6 to 12 months for proof of concept. Phase II supports prototype development with $400,000 to $1.8 million over 24 months. Phase III is where you commercialise the technology, but there’s no SBIR/STTR funding provided during this phase.
Here’s the hard truth. Only 16% of DoD SBIR-funded companies received Phase III contracts over the last decade. Fewer than 1% of Phase I awardees achieve Program of Record status.
SBIR awards are often called a “licence to hunt” because Phase I and II contracts don’t guarantee a long-term deal or path to large scale programs. But what they do is grant startups access to DoD stakeholders. You get to demonstrate customer demand and open doors to broader adoption.
Service innovation hubs like AFWERX and NavalX streamline the SBIR application process and accelerate evaluation timelines. They’re your entry point into the system.
Advanced computing and software leads cumulative investment at $90 billion from 2015 to first half 2025. Sensing, connectivity and security received $43 billion. Autonomous systems hit $26 billion. Space technology got $24 billion.
Most VC in defence has clustered around dual-use technologies – autonomy, AI-enabled decision systems, simulation, and sensing.
Dual-use technologies combine tangible deep-tech innovation with scalable software-led characteristics that investors find attractive. Shield AI, Skydio, and Applied Intuition were awarded OT prototyping contracts through DIU. The Hypersonix Launch Systems $46 million Series A round demonstrates how breakthrough propulsion technology can attract both government and private capital.
Dual-use enables revenue diversification. Defence contracts provide the stability. Commercial sales drive the volume and scale. Commercial markets typically offer faster procurement cycles – months instead of years – and lower regulatory overhead.
Defence validation creates commercial credibility. “If it’s good enough for the military” messaging resonates with enterprise buyers.
The challenges? ITAR restrictions limit international sales and technology transfer to commercial products. Successful dual-use requires separate product lines, distinct go-to-market strategies, and careful IP management.
Some examples: Cybersecurity firms turn defence-grade tech into enterprise zero-trust platforms. AI/ML companies convert intelligence analysis capabilities into predictive analytics products. Autonomous systems makers turn military drones into commercial inspection platforms.
The economics shift from government-dependent to commercially scalable, with defence as a strategic anchor customer.
U.S. scaleups like Shield AI, SpaceX, and Palantir demonstrated the efficacy of vertically integrated platforms initially focused on defence but now bridging the civil-military divide.
The Valley of Death is an 18 to 36 month funding gap. It sits between a successful SBIR Phase II prototype and a Phase III production contract. DoD’s PPBE budget cycle requires 2 to 3 years of planning to insert a new Program of Record. That creates this cash flow gap.
95-98% of SBIR Phase II recipients fail to achieve Program of Record status. Often it’s because of the Valley of Death.
STRATFI matches private investment dollar-for-dollar to fund transition activities. OSC loans provide below-market financing to extend runway. Prime contractor partnerships offer bridge funding in exchange for teaming agreements or acquisition options.
Breaking through requires understanding how DoD funds procurement at scale. Most successful startups combine multiple approaches – securing SBIR funding for credibility, using OTA contracts for rapid prototyping, partnering with a PEO to scale into a Program of Record.
DoD increasingly favours “buy and integrate” from startups over traditional in-house development.
Build advantages: custom fit to requirements, IP ownership, no vendor lock-in, higher security control. Buy advantages: faster deployment (12 to 18 months vs 3 to 5 years), lower upfront cost, proven technology.
The key evaluation factors are technology maturity, competitive advantage vs commodity, time-to-deployment urgency, and total cost of ownership. For a comprehensive framework on navigating these innovation opportunities and strategic trade-offs in the defense tech sector, see our complete guide.
A hybrid approach is emerging. Buy a commercial dual-use core, then customise the integration layer for classified applications. This gets you speed and cost efficiency while maintaining security requirements.
CMMC compliance, FedRAMP requirements, and ITAR restrictions apply whether you build or buy. OTAs facilitate access to new and commercial technologies outside standard government acquisition pathways.
Partnering with prime contractors – Lockheed Martin, RTX, Northrop Grumman, General Dynamics, Boeing – provides an entry point. Primes have established relationships, programs of record, contract vehicles, and deep institutional knowledge of military procurement.
Teaming with primes can offer much faster sales cycles for startups with products that align to existing programs.
The trade-off: subcontracting often means less control, lower margins, and longer sales cycles that limit your ability to drive change. But primes are massive customers constantly seeking innovative suppliers to integrate modern technology into defence programs.
The top 10 defense contractors retained about 65% market share across key segments over the past 10 years. That’s despite significant investment in new entrants.
Position yourself as an indispensable enabler. Leverage the primes’ scale while retaining a strategic path to long-term growth.
You’ll need FedRAMP authorisation for cloud-based services handling DoD data. CMMC Level 2 certification – that’s 110 controls – is required for contractors handling Controlled Unclassified Information. ITAR registration is necessary if you’re developing or exporting defence articles. And facility clearance is needed for classified work.
Timeline? 12 to 18 months total for the full compliance stack. Start with CMMC as the foundational requirement.
Foreign investment is permitted for unclassified dual-use technologies not subject to ITAR controls. But CFIUS review is required for foreign ownership exceeding 10% voting interest or 25% total equity.
UK, Australia, Canada – the Five Eyes partners – face fewer restrictions. Alternative approach: Foreign investors can participate through US-domiciled funds or as limited partners without board representation.
SBIR Phase I takes 6 to 12 months. Phase II takes 24 months. Then there’s the Valley of Death gap at 18 to 36 months. Program of Record insertion adds another 12 to 24 months. Total timeline: 5 to 8 years from initial SBIR award to sustained production revenue.
Only 2 to 5% of companies complete this journey. Acceleration is possible through OTA rapid prototyping – that’s a 2 to 3 year pathway – or prime contractor teaming.
OTA is a flexible contracting mechanism that’s exempt from Federal Acquisition Regulation (FAR) requirements. It enables rapid prototyping and production without the traditional procurement processes.
There are two types. Prototyping OTAs for technology demonstration. Production OTAs for follow-on manufacturing. Best used for dual-use commercial technologies, rapid iteration cycles, and non-traditional defence contractors.
You can access OTAs through the Defense Innovation Unit, service innovation hubs like AFWERX and NavalX, and OTA consortia like the National Security Innovation Network.
Europe has the National Security Strategic Investment Fund in the UK, the European Defence Fund that’s EU-wide, and individual country programs in France and Germany. Key differences: less restrictive export controls – there’s no ITAR equivalent – smaller individual fund sizes at $200M to $500M vs the multi-billion dollar US programs, and a stronger emphasis on NATO interoperability.
Advantages for US startups: dual-market strategy, less competition, faster procurement cycles. Challenges: currency risk, separate regulatory compliance, and relationship building required.
2 to 5% of SBIR Phase II recipients achieve Program of Record insertion. Of companies that enter the Valley of Death, approximately 15 to 20% successfully bridge to production contracts.
Success factors: a strong operational champion within the military service, demonstrated cost savings vs the incumbent solution, alignment with PPBE budget priorities, and bridge funding through STRATFI, OSC, or prime partnerships. Timeline averages 5 to 8 years from initial SBIR to sustained production revenue.
Depends on the classification level of the work. Unclassified work: no personal clearances required. Controlled Unclassified Information (CUI): no clearance needed but CMMC compliance is required. Secret or Top Secret work: facility clearance and employee clearances are mandatory.
Clearance timeline: 6 to 18 months depending on the level. Cost: $3,000 to $5,000 per employee for Secret, $10,000 plus for Top Secret. Many dual-use opportunities exist without classification requirements.
STRATFI – Strategic Financing – and TACFI – Tactical Financing – provide DoD matching funds to private investment. STRATFI handles larger awards, $5M to $50M, for strategic technologies and multi-year programs. TACFI provides smaller awards, $1M to $10M, for tactical solutions with faster deployment.
Both require a 1:1 private capital match, DoD end-user validation, and a pathway to Program of Record. Application is through service innovation hubs. AFWERX for Air Force STRATFI. NavalX for Navy programs.
PPBE – Planning, Programming, Budget, Execution – is DoD’s 2 to 3 year budget planning process. Planning year: define requirements and priorities. Programming year: allocate resources to programs. Budget year: Congressional appropriation. Execution: spend the appropriated funds.
Impact on you: a new Program of Record requires 2 to 3 years of PPBE insertion. This creates the Valley of Death. Your strategy should be to target mid-year requirements reviews, demonstrate cost savings to justify budget reallocation, and secure an operational champion to advocate in the planning cycle.
Yes, via the Defense Innovation Unit, service innovation hubs, and SBIR programs that are specifically designed for non-traditional contractors. Success factors: identify the dual-use application of your existing technology, partner with a military end-user for validation, and leverage your commercial traction as credibility.
Entry paths: DIU Commercial Solutions Opening (CSO) for rapid prototyping, SBIR for funded development, and OTA consortia for teaming opportunities. Avoid attempting traditional FAR-based procurement without experienced partners.
Loan structure: senior secured debt or subordinated convertible notes. Interest rates: below commercial rates, typically 3 to 6% vs 8 to 12% market rates. Repayment terms: milestone-based tied to contract awards, 3 to 7 year maturity.
Collateral: IP, contracts, equipment. Less stringent than commercial lenders. Conversion features: some loans are convertible to equity at future rounds. Amounts: $10M to $200M depending on company stage and private capital raised. The advantage? Non-dilutive capital extending your runway through the Valley of Death.
Map your technology to the DoD service and capability area. Army has 11 PEOs covering ground vehicles, aviation, missiles, soldier systems, and enterprise IT. Navy has 8 PEOs including ships, submarines, aircraft, and information warfare. Air Force has 7 PEOs for fighters, bombers, space, nuclear, and command and control.
Research the Future Years Defense Program (FYDP) for budget allocations. Attend industry days. Engage through the Defense Innovation Unit or service innovation hubs. Build relationships with PEO technical staff before any formal procurement starts.
Understanding the funding landscape is crucial, but it’s just one aspect of navigating the deep tech strategy shaping defense innovation in 2025. The intersection of government co-investment, venture capital, and breakthrough technologies creates unique opportunities for companies willing to engage with defense procurement complexity.
The 2025 defense tech investment environment offers unprecedented access to capital through hybrid funding models. Success requires understanding multiple funding pathways, navigating regulatory requirements, and positioning technology for both defense and commercial markets. Whether you’re evaluating strategic partnerships or build vs buy decisions, the key is aligning your technology roadmap with DoD procurement cycles while maintaining commercial viability.
Hypersonix Launch Systems – How an Australian Startup Is Building Hypersonic Aircraft with NASA and the PentagonBrisbane-based Hypersonix Launch Systems secured $46 million in Series A funding in 2024 to develop hydrogen-powered hypersonic aircraft. That’s the kind of money that gets deployed, not burned.
The founders have serious credentials. Dr. Michael Smart worked at NASA in the 1990s before leading hypersonic propulsion research at University of Queensland. He and David Waterhouse founded the company in 2019. They’re commercialising 35+ years of scramjet research and over 6,000 shock tunnel experiments.
The core technology is SPARTAN—a 3D-printed scramjet engine operating at Mach 5-12 with zero CO2 emissions. In 2024, the Pentagon selected Hypersonix for its HyCAT program from over 60 applicants. Their DART test vehicle launches from NASA Wallops in 2025.
The funding syndicate tells you this is more than a tech bet. Australia’s National Reconstruction Fund made its first defence investment. UK-based High Tor Capital led the round. Strategic investor Saab joined, along with QIC and Polish family office RKKVC.
This case study is part of our comprehensive guide to deep tech and defense innovation, where we explore how emerging technologies, strategic investment, and security considerations are reshaping the defense landscape. This is how deep tech startups bridge academic research to production, secure defence contracts, and structure multi-national funding rounds. Let’s get into it.
Hypersonix is commercialising scramjet research from University of Queensland. They’re the first company globally attempting commercial hydrogen-powered hypersonic aircraft.
The SPARTAN scramjet engine is fully 3D-printed. It’s air-breathing with no moving parts and can reach speeds of Mach 12. The company has 45 people in Brisbane developing dual-use technology for defence and commercial applications.
They’re building three products in sequence:
DART AE is a single-use demonstrator. Three-and-a-half metres long, flying Mach 5-7. This is the proof-of-concept.
VISR is the reusable ISR platform. Eight metres, Mach 5-10. This is where the business model lives—reusable intelligence, surveillance, and reconnaissance missions.
DELTA VELOS is the space launch vehicle. Sixteen metres, Mach 5-12. Think satellite launches and low Earth orbit supply runs.
Founded in 2019, first flight test in 2025. That’s a six-year development cycle to get a hypersonic vehicle airborne. Deep tech aerospace is not a sprint.
Dr. Smart’s background opened doors. They built HYPERTWIN X—a virtual design and testing environment leveraging all that experimental data—to reduce physical testing costs. When you’re pre-revenue and burning through a Series A, that matters.
The IP licensing from University of Queensland gave them a technical foundation. The challenge was translating lab prototypes to flight-ready hardware—manufacturing engineering and supply chain development you don’t get in a university lab.
A scramjet compresses incoming air through forward motion alone at hypersonic speeds. Combustion happens while air remains supersonic. No moving parts. Unlike turbojets with compressor blades, scramjets use geometry and speed for compression. For a deeper dive into the fundamentals of hydrogen-powered scramjet technology, including the physics behind Mach 12 flight capabilities, we’ve covered the technical details in a separate analysis.
They only work above Mach 5. This is why DART launches on a Rocket Lab HASTE booster to reach operational speed before the SPARTAN engine ignites.
The advantage? Air-breathing propulsion. Scramjets use atmospheric oxygen rather than carrying oxidiser like rockets. This dramatically reduces weight and cost for atmospheric flight.
The physics at hypersonic speeds is brutal. Temperatures exceed 3,000 degrees Fahrenheit and aerodynamic forces are extreme. You’re lighting fuel while air flows through the engine at Mach 5+ in milliseconds.
For reusable systems, scramjets make economic sense. They’re more fuel-efficient than rockets for atmospheric missions. This is why Hypersonix is building VISR as a reusable platform rather than disposable rockets.
SPARTAN burns hydrogen, not kerosene. It’s the only commercial scramjet using green hydrogen. Zero CO2 emissions, only water vapour.
Hydrogen offers higher specific impulse than kerosene. That means better performance at hypersonic speeds. The trade-off is cryogenic storage at -423°F. But for reusable hypersonic systems, the performance advantage outweighs the handling complexity.
The entire engine is additively manufactured. 3D printing enables complex internal geometries impossible with traditional machining. This includes cooling channels that manage extreme temperatures.
The design is modular. VISR uses four SPARTAN engines. This scalability means they can test propulsion at smaller scale before scaling up.
HYPERTWIN X gives them a development advantage. It’s a virtual environment built on decades of experimental data. This reduces physical testing costs and accelerates iteration. When you’re pre-revenue and burning through a Series A, that matters.
Defence organisations are increasingly interested in green capabilities. Zero emissions gives you a talking point that kerosene scramjets don’t have. It might sound like marketing, but governments care about this stuff.
Dr. Smart’s path was NASA researcher in the 1990s, then University of Queensland Chair of Hypersonic Propulsion, then startup CTO in 2019. Hypersonix represents the culmination of decades of research into scramjet propulsion.
The IP came from University of Queensland. Thirty-five years of scramjet research. Six thousand shock tunnel experiments. Australia has been a global leader in hypersonic technology since 1989.
The challenges? Translating lab prototypes to flight-ready systems. Moving from Technology Readiness Level 3-4 to TRL 6-7 requires manufacturing engineering you don’t develop in a university lab. You need production processes, supply chains, quality control.
Hypersonic engineers are scarce globally. You’re competing with Lockheed, Boeing, and every other defence prime for talent. And you’re doing it from Brisbane.
HYPERTWIN X keeps development costs manageable. Virtual testing means fewer expensive flight tests. This is how you manage cash when you’re pre-revenue.
The lesson for deep tech founders? Technical credibility matters. Smart had NASA and University of Queensland on his resume. That opens doors. It gets you meetings with defence procurement officers and venture capitalists who understand the space.
HyCAT stands for Hypersonic and High-Cadence Airborne Testing Capabilities. It’s a US Department of Defense program managed by Defense Innovation Unit. The goal is dramatically increasing test frequency while lowering costs.
Right now, hypersonic tests cost about $100 million per flight and happen once or twice per year. The Pentagon wants to conduct up to 50 flight tests annually. That’s a 25-50x increase in test cadence.
Hypersonix was the first company selected from more than 60 applicants. Here’s why they won:
Only hydrogen scramjet applicant—differentiation matters in defence procurement. Everyone else is using kerosene.
Technical heritage from University of Queensland and NASA—credibility with evaluators who know the field.
Australian AUKUS partner status—geopolitical alignment matters when you’re dealing with the Pentagon.
3D printing reduces costs—lower cost per test than traditional approaches.
Realistic development timeline—they didn’t overpromise. Six years to first flight is honest. Defence has been burned too many times by optimistic schedules.
Defense Innovation Unit focuses on accelerating commercial technology adoption. DIU program manager Maj. Ryan Weed described the effort as a “paradigm shift, viewing the hypersonic realm as a place for aircraft, not just missiles and weapons”.
The contract includes DART AE test flight at NASA Wallops using Rocket Lab’s HASTE booster in 2025. That’s the proof point.
Winning HyCAT gave them Department of Defense validation. That matters for follow-on production contracts. DoD doesn’t hand out contracts to 45-person startups unless the technical evaluation is solid.
High Tor Capital led the round. The syndicate included Australia’s National Reconstruction Fund ($10 million), Queensland Investment Corporation, strategic investor Saab, and Polish family office RKKVC.
The NRFC investment marked their first defence sector allocation. NRFC is Australia’s sovereign investor with $15 billion to deploy. Getting their first defence investment sends a signal.
The multi-national syndicate spanned UK, Australia, Sweden, and Poland. That’s validation across allied nations and market access through investor networks.
Strategic investor Saab brings defence industry expertise, customer relationships, and partnership opportunities. In defence tech, choosing investors who understand procurement timelines and customer dynamics matters more than pure capital.
North Ridge Partners acted as financial advisor with aerospace and defence expertise. Complex cross-border defence transactions need advisors who understand export controls, ITAR compliance, and international defence partnerships.
David Gall, NRFC CEO, noted they “see huge potential in backing Australian companies and innovations that build our sovereign capability”.
Series A timing was deliberate. Post-HyCAT selection but pre-flight test. That’s validation without full technical de-risking. The funding supports DART flight testing, VISR development, and manufacturing scale-up.
Defence tech funding differs from SaaS. Timelines are longer—expect 18+ months versus 6 months for a typical software raise. Strategic investors matter more than your cap table looking good on Twitter. Sovereign funds get involved. Export controls complicate term sheets. Valuation is based on IP strength, team pedigree, and strategic importance, not ARR multiples. For a comprehensive overview of defense tech investment trends and how government-venture capital partnerships are reshaping the funding landscape, we’ve analysed the broader patterns driving this $46 million round.
Supersonic combustion is the first problem. Achieving stable combustion while air flows through the engine at Mach 5+ in milliseconds. It’s been described as lighting a match in a hurricane. That’s not marketing hyperbole.
Thermal management comes next. At Mach 10, temperatures exceed 1,800 degrees Celsius. The airframe needs advanced materials like ceramic matrix composites. Standard aerospace alloys melt.
Hydrogen fuel handling adds operational complexity. Cryogenic storage at -423°F requires specialised infrastructure. You can’t just fill up at any airfield.
Reusability engineering separates DART from VISR. DART is single-use. VISR must survive multiple flights with thermal protection systems, structural fatigue management, and maintenance protocols. This is the difference between a technology demonstrator and a business.
Manufacturing complexity doesn’t end after the first unit. 3D printing high-temperature alloys at scale. Quality control for safety-critical components. Supply chain for exotic materials. Aerospace manufacturing is hard; hypersonic aerospace manufacturing is harder.
SPARTAN’s 3D-printed cooling channels address thermal management. HYPERTWIN X simulation reduces physical testing needs. The modular design enables incremental development.
But these vehicles travel at speeds exceeding Mach 5, creating physics challenges that computer models alone cannot fully replicate. You need real-world flight data. There’s no getting around it.
DART AE launches in 2025. It’s a 3.5-metre single-use vehicle flying Mach 5-7. It’s the world’s first 3D-printed hypersonic airframe. This is the technology demonstrator.
VISR comes next, targeting 2027-2028. Eight metres long, reusable, Mach 5-10, powered by four SPARTAN engines. Designed for ISR missions and rapid payload delivery. This is the revenue vehicle—where the business model kicks in.
DELTA VELOS is the long-term play. Sixteen metres, Mach 5-12, for satellite launches and low Earth orbit supply missions. This is years away but it’s where the big contracts live.
The revenue strategy starts with defence contracts for ISR missions. Then commercial satellite launch services. Then hypersonic testing platform services for aerospace customers who can’t afford $100 million per test.
The milestone-based approach manages risk. DART success unlocks follow-on HyCAT contracts. VISR demonstration enables production orders. Each step de-risks the next.
NRFC funding pays for product development and establishment of advanced manufacturing capabilities in Queensland. That’s jobs and sovereign capability, which governments care about.
Six years from founding to first flight test. You need patient capital and technical credibility. There are no shortcuts in aerospace. Anyone promising faster timelines is lying to you or themselves.
AUKUS is the Australia-UK-US security partnership. Hypersonic capabilities are prioritised under AUKUS Pillar II. Australia’s National Defence Strategy 2024 identified hypersonic capabilities as a key defence priority with $3 billion budget over the next decade.
AUKUS streamlines export controls. ITAR compliance is easier for allied companies. Technology transfer moves faster. This matters when you’re trying to work with NASA and the Pentagon from Brisbane.
The US hypersonic investment is measured in billions annually. HyCAT is an entry point. AUKUS status positions Hypersonix for larger follow-on production contracts that non-allied companies simply can’t access.
The market access pathway: HyCAT demonstration proves technology, then production ISR contracts, then integration with US platforms, then broader allied nation sales. Each step builds on the previous one.
NRFC CEO David Gall sees potential in “tapping into the global market for hypersonic and counter-hypersonic technologies among our allies”.
Being an AUKUS ally matters when you’re a 45-person startup trying to sell into the Pentagon. It’s the difference between getting meetings and getting ghosted. Without it, you’re competing on a level playing field with Chinese companies, and good luck with that.
Supersonic means faster than sound—Mach 1+, or 767+ mph. Hypersonic is Mach 5+, around 3,800+ mph. But the distinction isn’t just speed.
The physics changes. Hypersonic speeds create extreme heating from atmospheric compression. The air behaves differently. You need air-breathing scramjet engines instead of traditional turbojets because turbojets can’t handle the airflow speeds.
It’s not just “faster supersonic.” It’s a different engineering problem entirely.
Hydrogen offers higher specific impulse than kerosene—better performance at hypersonic speeds. Zero CO2 emissions, only water vapour. The trade-off is cryogenic storage at -423°F.
For reusable systems optimising for performance and sustainability, hydrogen wins. Hypersonix is the only commercial scramjet developer betting on hydrogen. Everyone else is using kerosene.
Is it harder to handle? Yes. Does it perform better and align with government sustainability priorities? Also yes. That’s the bet they’re making.
At Mach 7, the Sydney-to-London distance would take roughly 2 hours versus 22+ hours for current commercial flights.
But DART and VISR are military ISR platforms, not passenger aircraft. Hypersonic passenger flight remains a future possibility, not a near-term commercial product. Don’t expect to book a ticket anytime soon.
SPARTAN is at TRL 5-6 currently. The DART flight test in 2025 advances it to TRL 6-7. Full TRL 9 requires successful VISR flights and sustained operational use.
The 6,000+ shock tunnel experiments at University of Queensland accelerated early development. That’s years of testing data they didn’t have to generate from scratch. It’s also why they could credibly promise a 2025 flight test to the Pentagon.
Scramjets only function above Mach 5. DART launches on a Rocket Lab HASTE booster to reach operational speed, then SPARTAN ignites.
VISR plans runway takeoff using conventional jet engines to accelerate, then scramjet activation at hypersonic speeds. This combined cycle approach is standard. You need something to get you to Mach 5 before the scramjet can take over.
Think of it like a two-stage system. You don’t try to use the scramjet until conditions are right.
Additive manufacturing eliminates expensive tooling and moulds. It enables rapid iteration—days versus months for design changes.
You can create complex internal geometries impossible to machine, like SPARTAN’s cooling channels. Traditional machining can’t create the internal cooling passages they need for thermal management.
For low-volume, high-complexity aerospace applications, 3D printing offers substantial cost reduction. You’re not building 10,000 units. You’re building tens. The economics are completely different.
VISR is 8 metres long with four SPARTAN engines. Specific payload capacity hasn’t been disclosed publicly.
Industry-standard hypersonic ISR platforms carry 200-500 kg. But the value proposition isn’t payload mass—it’s speed and survivability. Mach 5-10 enables rapid global response within 1-2 hours. That’s the capability governments are paying for.
Hypersonix has approximately 45 employees in Brisbane. The team includes aerospace engineers, materials scientists, manufacturing engineers for 3D printing, and business development staff.
Key challenge: hypersonic expertise is scarce globally. You’re competing with Lockheed, Boeing, Northrop Grumman, and every other defence prime for the same small pool of talent.
Being in Brisbane rather than Los Angeles or Washington DC makes hiring interesting. You’re selling lifestyle and opportunity over brand name.
SPARTAN engines produce zero CO2 emissions. Only water vapour when using green hydrogen from renewable energy electrolysis. Kerosene scramjets emit CO2.
Reusable systems like VISR reduce per-mission environmental impact. But life-cycle analysis must include hydrogen production source. Green hydrogen from renewables is essential for true zero-carbon operation.
If you’re producing the hydrogen from natural gas, you’re just moving the emissions somewhere else. That’s why “green hydrogen” matters—it’s hydrogen produced using renewable electricity.
DART AE is scheduled for launch from NASA Wallops Flight Facility in Virginia in 2025 using a Rocket Lab HASTE booster. Specific date depends on final integration milestones.
Official updates come through Hypersonix Launch Systems’ website, LinkedIn, and Twitter. NASA and Defense Innovation Unit may provide additional coverage. Defence programs sometimes go quiet for security reasons, so don’t expect real-time updates like a SpaceX launch.
The strategy focuses on commercial innovation advantages. 3D printing enables faster iteration and lower costs. Hydrogen propulsion differentiates from kerosene competitors. Small team moves faster than large bureaucracies.
AUKUS partnership provides market access. Rather than competing for large prime contracts initially, Hypersonix targets niche programs like HyCAT where innovation matters more than scale. Then expand once technology is proven.
Strategic investor Saab may facilitate partnerships with larger primes. This is common in defence—small companies develop breakthrough technology, then partner with big players for production and integration.
Competitors include Hermeus in the US pursuing kerosene scramjet targeting Mach 5 passenger aircraft. Venus Aerospace in the US developing rotating detonation engine with Mach 9 concepts. Destinus in Switzerland working on hydrogen-powered hypersonic freight.
Traditional defence primes like Lockheed Martin, Boeing, Northrop Grumman, and Raytheon are developing military hypersonic weapons but focused on missiles, not reusable aircraft.
Hypersonix differentiates through hydrogen fuel, reusable ISR platform focus, and AUKUS allied nation status. The hydrogen bet is unique. Everyone else is going kerosene.
Hypersonix demonstrates how deep tech startups navigate the complexity of defence innovation—from translating academic research into production-ready systems to structuring multi-national funding rounds and securing Pentagon contracts. The DART flight in 2025 will validate whether this approach delivers on the promise. For comprehensive strategic lessons on defense innovation, including investment patterns, security considerations, and technology trends across the sector, see our complete guide to deep tech and defense innovation.
How Hydrogen-Powered Scramjets Are Enabling Mach 12 FlightSustained hypersonic flight above Mach 5 has been a hard aerospace engineering problem for decades. You can strap a rocket to something and push it to hypersonic speeds, but rockets carry their oxidiser onboard. That limits both range and flight duration. Air-breathing engines solve this by compressing atmospheric oxygen, enabling extended hypersonic operations.
Scramjet engines—supersonic combustion ramjets—take this to the extreme. They work at Mach 5 and beyond, with no moving parts. Hydrogen fuel gives you energy density that kerosene can’t touch (142 MJ/kg versus 43 MJ/kg) and produces zero CO2 when it burns.
This article explores the technical foundations of scramjet propulsion as part of our broader analysis of deep tech and defense innovation.
Recent progress in materials science, 3D printing, and thermal management has pushed the envelope to Mach 12. Australian startup Hypersonix raised $46 million in Series A funding in October 2025 to advance their SPARTAN engine, which demonstrates hydrogen-powered scramjet technology reaching these speeds. The engine design draws on over 6,000 shock tunnel experiments conducted at the University of Queensland.
Understanding how these systems work—the physics, the materials choices, the manufacturing approaches—gives you insight into where aerospace innovation is heading.
A scramjet compresses incoming air at hypersonic speeds using only aerodynamic forces. No compressor blades, no turbines, no moving parts. The vehicle’s speed does the compression work.
Three sections make up the engine: an inlet that compresses air through shock waves, a combustion chamber where fuel burns in supersonic airflow, and a nozzle that accelerates the exhaust to produce thrust.
Air enters at Mach 5 to 12, stays supersonic—typically Mach 2 to 3—in the combustion chamber. Fuel injection and combustion happen in under a millisecond.
The key difference from other jet engines: scramjets only work at hypersonic speeds. They need a rocket booster or carrier aircraft to reach operational velocity.
Think of it as a flying stovepipe. Air enters, compresses, burns, exits.
A ramjet slows incoming air to subsonic speeds before combustion. That works from Mach 3 to 6, but beyond Mach 6 the pressure losses destroy efficiency. Scramjets avoid this by maintaining supersonic flow throughout.
Hypersonix’s SPARTAN engine demonstrates this in hardware. The Australian startup’s approach combines inlet geometry validated through extensive ground testing with combustion chamber design optimised for hydrogen fuel.
Regular combustion in subsonic engines allows fuel and air to mix over seconds or minutes. Supersonic combustion completes in under a millisecond.
In a typical jet engine, air slows to near-zero velocity. Scramjets maintain Mach 2 to 3 airflow through the combustor. The supersonic flow prevents flame from travelling upstream—a phenomenon called flashback that would destroy the engine.
Fuel injector design becomes critical. Hydrogen must penetrate supersonic crossflow and mix rapidly. The extreme temperature and pressure causes autoignition—no spark plugs needed. But the short residence time limits fuel burning. Combustion efficiency typically hits 70 to 85 per cent, compared to 98 per cent in subsonic engines.
Thermal management adds complexity. Inlet temperatures reach 1,000 to 1,500 degrees Celsius. Peak flame temperatures hit 2,000 to 2,500 degrees.
NASA is developing cavity flame holder technology to reduce combustor length by 25 per cent. These are recessed pockets in the combustor wall that create subsonic recirculation regions where fuel can ignite, then spread to the main flow.
The physics here matter. Flame speed relative to flow velocity determines whether combustion is even possible. Hydrogen’s fast flame speed—2 to 3 metres per second versus 0.4 metres per second for kerosene—enables reliable ignition in millisecond timeframes.
Hydrogen gives you the highest specific energy of any chemical fuel: 142 MJ/kg versus 43 MJ/kg for kerosene. That’s more than three times the energy density by weight.
Faster flame speed matters at hypersonic residence times. Hydrogen combustion produces only water vapour—zero CO2, zero NOx at controlled temperatures, zero particulates. The lower molecular weight of the exhaust gases gives you higher specific impulse.
Liquid hydrogen at minus 253 degrees Celsius offers excellent cooling properties. It can absorb heat from engine components before combustion, implementing regenerative cooling that recovers waste heat for thrust.
Hydrogen also has a wider flammability range—4 to 75 per cent in air versus 0.6 to 5.5 per cent for kerosene. This improves combustion stability.
The challenges are real though. Cryogenic storage adds complexity. Hydrogen needs larger volume tanks. Hydrogen embrittlement affects material selection—hydrogen diffuses into metals, causing brittleness.
Unlike conventional scramjets that rely on kerosene, Hypersonix’s SPARTAN scramjets use hydrogen, producing zero carbon emissions. The DART AE demonstrator aims to achieve the first sustained hypersonic flight using green hydrogen.
Green hydrogen production via electrolysis using renewable electricity creates a carbon-neutral fuel cycle. This makes sustainable hypersonics feasible.
Aerodynamic heating at Mach 12 generates surface temperatures exceeding 1,800 degrees Celsius. Temperature rises with the square of velocity—the jump from Mach 5 to Mach 12 represents a 5.8-fold temperature increase.
The combustion chamber experiences combined heating: compressed air enters at 1,000 to 1,500 degrees, then combustion adds peak flame temperatures of 2,000 to 2,500 degrees. The surface might be at 1,800 degrees while internal structure must stay below material limits.
Engine components face cyclic thermal loads—heating during powered flight, cooling during coast phases.
At Mach 10, temperatures exceed 1,800 degrees Celsius, so you need advanced materials like ceramic matrix composites. Material temperature limits constrain design. Aluminium alloys max out at 300 degrees. Titanium handles 600 degrees. Nickel superalloys reach 1,000 degrees. Ceramics can withstand 1,800 degrees and beyond.
The design trade-offs are straightforward: add cooling mass, use higher-temperature materials, or limit flight duration. Each choice affects payload capacity, cost, and operational flexibility.
Additive manufacturing produces complex internal geometries that traditional machining cannot. Integral cooling channels, optimised flow paths, biomimetic designs—these are all possible with 3D printing.
Design-to-hardware time drops from 12 to 18 months to 4 to 6 weeks. Direct metal laser sintering works with high-temperature alloys like Inconel 718 and Inconel 625, which withstand 700 to 1,000 degrees.
Multiple machined components become a single printed assembly, cutting weight and potential failure points.
The SPARTAN engine’s 3D-printed design demonstrates specific capabilities: fuel injector struts with internal cooling channels, combustion chambers with integrated features.
The process: CAD design, CFD optimisation, DMLS printing, heat treatment, machining, inspection.
GE Aviation, SpaceX, and Relativity Space also use additive manufacturing for propulsion components. Hypersonix’s approach takes this further—vertical integration of design, printing, and testing enables rapid development cycles.
Ceramic matrix composites operate at 1,400 to 1,800 degrees Celsius. Silicon carbide fibres in a silicon carbide matrix—SiC/SiC—go where metals can’t.
Refractory metals handle extreme zones. Tungsten alloys melt at 3,400 degrees. Molybdenum and niobium work for leading edges and combustor sections.
Nickel-based superalloys like Inconel cover the 700 to 1,200 degree range.
Carbon-carbon composites reach 2,000 degrees and beyond. The Space Shuttle used them.
Thermal barrier coatings provide 100 to 200 degrees of temperature reduction.
The trade-offs: Ceramics are brittle. Refractory metals are dense. Superalloys have temperature limits.
VISR will be built using high-temperature ceramic composites, demonstrating how these materials enable operational vehicles. CMCs and refractory metals run 10 to 100 times more expensive than titanium or aluminium. These material choices constrain where hypersonic technology can be economically deployed—which is why defense tech investment increasingly focuses on breakthrough propulsion technologies.
Ramjets slow incoming air to subsonic speeds—Mach 0.3 to 0.5—before combustion. This works efficiently from Mach 3 to 6.
Scramjets maintain supersonic airflow throughout. The combustion happens at Mach 2 to 3. This extends operational range to Mach 5 to 15 and beyond.
The transition happens because slowing Mach 6 and higher air to subsonic creates excessive pressure losses and temperatures. Scramjets avoid this by burning fuel in supersonic flow, accepting the combustion challenges this creates.
SPARTAN uses hydrogen fuel with 142 MJ/kg energy density and fast flame speed. The 3D-printed Inconel combustion chamber integrates cooling channels. Inlet geometry was validated through extensive shock tunnel experiments.
The design balances inlet compression, combustion efficiency, thermal management, and nozzle expansion. Development used the HYPERTWIN X virtual testing environment, combining CFD with shock tunnel data.
No. Scramjets need hypersonic speeds—Mach 4 to 5 minimum—to generate inlet compression. Vehicles need rocket boosters or carrier aircraft launch to reach operational velocity.
The DART AE will launch aboard Rocket Lab’s HASTE booster from NASA’s Wallops Flight Facility. This limitation restricts scramjets to specific applications: hypersonic missiles, research vehicles, space launch upper stages, reconnaissance platforms.
Thermal loads at Mach 12 require expensive ceramic materials. Supersonic combustion instabilities and low efficiency—70 to 85 per cent fuel burn—limit performance.
Hydrogen storage complexity adds operational burden. Cryogenic tanks, boil-off management, and safety protocols complicate operations. Materials degradation from oxidation, thermal fatigue, and hydrogen embrittlement affects service life.
Infrastructure doesn’t exist yet. Hypersonic test facilities are rare. Hydrogen refuelling at operational scale hasn’t been deployed.
Traditional full-up flight tests cost about $100 million per flight. Cost per flight runs thousands of times higher than subsonic aircraft.
Both use similar metal additive processes. But hypersonic engines face sustained thermal exposure measured in minutes. Rockets see impulse heating measured in seconds.
Scramjets need complex internal flow paths for inlet compression and fuel mixing. Rockets optimise for throat and expansion geometry.
Hypersonic engines integrate cooling channels more extensively. Material choices overlap—Inconel, copper alloys—but scramjets use ceramic matrix composite inserts where rockets rely on ablative cooling.
CFD simulates hypersonic airflow, shock interactions, and combustion. This predicts performance before expensive hardware testing.
Hypersonix’s HYPERTWIN X environment combines CFD with shock tunnel experimental data. Benefits: reduce physical test count, explore broader design spaces. Shock tunnel tests cost $10,000 to $50,000 each.
Limitations: Turbulence models remain imperfect at hypersonic conditions. Validation against experimental data is required.
Liquid hydrogen at minus 253 degrees Celsius needs cryogenic tanks with multi-layer vacuum insulation.
Challenges: continuous boil-off requiring venting or active cooling. Large volume requirements—hydrogen is four times less dense than kerosene. Hydrogen embrittlement as hydrogen diffuses into metals. Leaks create explosion risk.
VISR and DELTA VELOS integrate conformal tanks within the airframe. Green hydrogen production via renewable electrolysis enables sustainable operations.
Specific impulse measures propulsion efficiency: seconds of thrust per unit weight of propellant. Higher specific impulse means greater range or payload capacity.
Hydrogen scramjets achieve 1,500 to 2,500 seconds at Mach 7 to 12. Kerosene ramjets reach 800 to 1,200 seconds. Rockets deliver 280 to 450 seconds.
The air-breathing advantage: not carrying oxidiser onboard improves efficiency. Scramjets excel at sustained hypersonic cruise. Rockets are better for acceleration and exo-atmospheric flight.
Current technology demonstrates 5 to 10 minutes continuous operation. The X-51 Waverider achieved 6 minutes at Mach 5.
Limitations: fuel capacity, thermal soak, engine durability, trajectory constraints. The SPARTAN engine is designed for multiple burn cycles on reusable vehicles.
Future goal: 30 to 60 minute hypersonic cruise. Thermal management and structural fatigue are the primary limiters.
UQ operated shock tunnel facilities that conducted over 6,000 hypersonic experiments. These generated foundational data for SPARTAN engine design.
Shock tunnels create milliseconds of Mach 5 to 12 flow conditions. This enables ground testing of inlet performance, combustion ignition, and thermal loads.
Hypersonix’s founding team includes UQ researchers. Australia has been a global leader in hypersonic technology since 1989.
Mach 12 equals 9,200 mph. Low Earth orbit velocity is 17,500 mph. Scramjets alone can’t reach orbital speed but reduce rocket propellant needed.
Two-stage-to-orbit: scramjet accelerates to Mach 10 to 12 at 100,000 feet. Rocket completes orbital insertion. Benefit: 30 to 40 per cent propellant mass reduction.
DELTA VELOS is designed for this profile, carrying small satellites to orbit. The challenge: air density at 100,000 feet becomes too low for scramjet thrust.
Hydrogen combustion produces only water vapour: H2 plus O2 yields H2O. Zero CO2 emissions.
Scramjets produce some thermal NOx, but far less than kerosene. No particulates, no soot, no unburned hydrocarbons.
Green hydrogen pathway: electrolysis using renewable electricity creates a carbon-neutral fuel cycle.
Contrails from water vapour create climate impact. Ice crystals reflect sunlight. Lower flight paths can eliminate this.
Hydrogen-powered scramjet technology represents a convergence of materials science, propulsion physics, and manufacturing innovation. As development costs and complexity drive collaboration between startups, defense agencies, and investors, understanding these technical fundamentals becomes essential for evaluating opportunities in the broader deep tech and defense innovation landscape.
Deep Tech and Defense Innovation – Opportunities, Risks and Strategic Lessons from 2025Defense innovation is shifting from prime contractor monopolies to startup-driven breakthrough technologies. Hydrogen-powered hypersonic aircraft are transitioning from research labs to Pentagon testing programs. Government funds are partnering with venture capital to accelerate startup innovation. Australian companies are competing globally in advanced propulsion systems. High-profile insider threat cases remind us that strategic advantages come with security responsibilities.
This comprehensive guide synthesises lessons from hypersonic technology development, startup success stories, investment patterns, and cybersecurity incidents into a strategic framework for evaluating defense opportunities. It provides overview-level context on the technologies, ecosystems, and decisions shaping this landscape, with clear paths to deeper technical and analytical resources.
What you’ll explore here:
Think of this as your strategic entry point. The detailed technical explanations, case studies, and implementation guidance live in those linked articles. This page provides the landscape view that helps you decide where to dive deeper.
Deep tech refers to innovations based on substantial scientific or engineering advances rather than software-only solutions. Unlike SaaS platforms, deep tech typically requires significant R&D investment, longer development timelines, and hardware-intensive infrastructure. In defense contexts, this includes hypersonic propulsion, autonomous platforms, advanced materials, and quantum technologies—innovations with dual-use applications across commercial and military domains. Evaluating these opportunities requires careful assessment of technical complexity and capital requirements.
Software startups typically reach product-market fit in 6-18 months with $500K-2M seed capital. In contrast, deep tech startups require 5-7 years and $40-50M to reach first operational prototype, spending years developing technologies before their first flight or deployment. This capital intensity historically limited defense innovation to large prime contractors with patient balance sheets.
Dual-use technologies are changing this dynamic. They attract venture capital comfortable with longer timelines but seeking larger addressable markets. Autonomy, AI decision systems, and hypersonic propulsion address real-world challenges while leveraging software-intensive growth dynamics that venture investors understand. This convergence of deep tech foundations with software-enabled scale explains why defense innovation is attracting unprecedented venture capital. US scaleups like Anduril and Palantir have demonstrated that integrated platforms can bridge the civil-military divide profitably.
Learn more: How Hydrogen-Powered Scramjets Are Enabling Mach 12 Flight explores hypersonic propulsion fundamentals, while Hypersonix’s startup journey shows dual-use innovation in practice.
Defense breakthroughs in 2025 span hypersonic propulsion, autonomous systems, advanced manufacturing, and electronic warfare. Hypersonic capabilities—sustained flight above Mach 5—enable rapid response and evade traditional defense systems. These technologies enable strategic advantages: hydrogen-powered scramjet engines provide reusable hypersonic flight, 3D-printed components withstand temperatures exceeding 1,800°C, and AI-enabled autonomous platforms transition from research to operational testing. Government-private partnerships and strategic competition with China and Russia accelerate this development.
The SPARTAN scramjet engine exemplifies these advances: fully 3D-printed, air-breathing, reaching Mach 12 with zero emissions. Unlike kerosene-based scramjets, hydrogen systems enable reusable hypersonic flight. This represents a fundamental change from single-use missiles to reusable aircraft—aircraft that can be tested, refined, and deployed repeatedly rather than expended on a single mission. Key enablers include ceramic composites surviving extreme temperatures, computational simulation reducing physical testing costs, and additive manufacturing enabling rapid prototyping of complex geometries.
The Pentagon is reconsidering its dependence on high-value platforms like aircraft carriers and stealth fighters, pursuing hybrid approaches that integrate smaller, affordable autonomous systems alongside traditional assets. This shift signals a rethinking of military doctrine toward distributed, technology-intensive force structures. China’s hypersonic glide vehicles and Russia’s Avangard and Kinzhal systems drive Western urgency in developing comparable capabilities.
Explore further: Our technical deep-dive on how scramjet technology works explains scramjet physics and engineering challenges. Hypersonix’s case study shows how one Brisbane startup is implementing these technologies.
Australia is establishing sovereign hypersonic capabilities through strategic government investment, world-class research infrastructure, and startup innovation. Hypersonix Launch Systems exemplifies this approach: University of Queensland foundations (30+ years of research, 6,000+ shock tunnel experiments), $46 million Series A from international investors including the National Reconstruction Fund Corporation (NRFC), and participation in the Pentagon’s HyCAT testing programme. The AUKUS partnership amplifies Australia’s strategic relevance, creating technology transfer opportunities that position Brisbane as an emerging deep tech hub.
NRFC’s $10 million equity investment represented a strategic shift for the fund. Before Hypersonix, NRFC focused primarily on advanced manufacturing and renewable energy sectors. As CEO David Gall explained: “Defense is one of our priority areas. We see huge potential backing Australian companies building sovereign capability while tapping into global markets for hypersonic technologies among our allies.”
The staged product roadmap illustrates the sovereign capability approach: DART AE (3.5-metre testing vehicle, Mach 5-7), VISR (8-metre reusable aircraft, Mach 5-10), and DELTA VELOS (16-metre, Mach 5-12). AUKUS creates market access well beyond Australia’s modest defense budget of roughly $50 billion annually. The partnership provides pathways to US and UK procurement budgets totalling over $1 trillion, justifying venture-scale capital deployment in Australian defense tech. The HYPERTWIN X virtual simulation environment, drawing on decades of University of Queensland experimental data, enables rapid iteration without the cost of physical testing for each design change.
Full details: The Hypersonix success story covers the journey from university research to Pentagon testing, including founder Dr Michael Smart’s NASA experience and the technical breakthrough of hydrogen-powered scramjets.
Traditional defense procurement timelines and venture capital exit horizons historically misaligned, creating a “Valley of Death” where startups exhausted capital before securing production contracts. Sales cycles stretch years through approval processes and qualification regimes. Startups face 7-10 year fund horizons and growth milestones misaligned with multi-year procurement timelines. Government co-investment models—Australia’s NRFC, US Office of Strategic Capital, NATO Innovation Fund—bridge this gap with patient capital, de-risking private investment and accelerating procurement pathways. This unlocks startup agility while maintaining strategic control over critical capabilities.
Government participation in funding rounds sends powerful market signals. It validates strategic importance, technical feasibility, and procurement likelihood—reducing risk perceptions that traditionally kept venture capital away from defense hardware. When NRFC joined Hypersonix’s Series A alongside High Tor Capital (UK-based national security VC), QIC (Queensland sovereign fund), Saab (Swedish defense prime), and RKKVC (Polish government fund), the mixed investor base demonstrated the new co-investment model working across borders.
This model is becoming standard across allied nations. NATO’s 32 members pledged 5% GDP defense spending by 2035, with portions explicitly designated for startup innovation programmes. The Defense Innovation Unit’s Replicator programme channelled $500 million to nontraditional competitors for rapid prototyping of autonomous capabilities. Governments gain startup speed and commercial innovation; venture investors gain patient capital co-investors and procurement pathway visibility; talent gravitates toward mission-driven dual-use businesses offering both impact and exit potential.
Investment analysis: Our defense tech investment overview examines co-investment models, investor motivations, and build-buy-partner frameworks in detail.
Defense tech investment concentrates in autonomous systems, hypersonic capabilities, cybersecurity, and space technologies. Through H1 2025, US defense tech startups raised $38 billion—potentially exceeding the 2021 peak of $42 billion if the pace continues. Beyond aggregate volume, three patterns define the landscape: mega-rounds demonstrate confidence (Anduril’s $1.5 billion Series F led by Founders Fund), early-stage capital flows to emerging categories (counter-drone systems, electronic warfare), and international diversification as NATO and Asia-Pacific investors enter the market. Investor types include national security-focused VCs (High Tor Capital, Point72 Ventures), strategic corporates (Saab, BAE Systems), sovereign wealth funds (QIC), and traditional VCs entering via “American Dynamism” strategies.
The economics attract capital previously reserved for software. The US spent $997 billion on military procurement and R&D last year. Capturing even 1-2% of addressable spending represents substantial opportunities with sticky, recurring revenue streams similar to enterprise SaaS contracts. Defense Innovation Unit programmes provide non-dilutive funding through SBIR/STTR grants while accelerating procurement pathways. The challenge: converting prototype contracts into programs of record with multi-year production commitments.
Geographic patterns are shifting. US dominance continues (roughly 85% of defense tech venture capital), but the NATO Innovation Fund’s €1 billion commitment and NRFC’s defense mandate signal recognition that sovereign capability requires venture-scale risk capital. Exit considerations matter critically. The top 10 contractors retained approximately 65% market share despite new entrant investment—exits happen via program of record adoption, strategic acquisition by primes, or public markets (following Palantir’s successful IPO). By 2030, new entrants need $15-20 billion in aggregate revenues (5-7% of addressable procurement spend) to justify current valuations—requiring genuine capability delivery, not just prototype demonstrations.
Recent examples beyond Anduril: Shield AI raised $300 million for autonomous flight software, Vannevar Labs raised $150 million for defense data platforms, and Rebellion Defense raised $150 million for AI-enabled defense applications. The capital is available; the question is whether entrants can navigate from prototype to production at scale.
Complete breakdown: Defense tech investment trends and the funding landscape provides detailed investor profiles, investment frameworks, and exit pathway analysis.
Technology companies can engage defense innovation through multiple pathways: developing dual-use technologies applicable to military needs, partnering with defense startups as strategic investors or technology providers, competing for government innovation programmes (DIU, AFWERX), or pursuing build-versus-buy evaluations for acquiring defense capabilities. Opportunities span autonomous systems, cybersecurity, data analytics, communications infrastructure, and advanced manufacturing. Success requires understanding procurement processes, navigating export controls, and evaluating alignment with organisational values and risk tolerance.
The DoD designates 14 critical technology areas including quantum science, AI and autonomy, space technology, and advanced materials. Current Undersecretary Emil Michael is streamlining this to focus on fewer, higher-impact areas through “sprints” delivering technology to armed forces within 2-3 years rather than the traditional decade-plus timelines. Dual-use technologies leverage familiar software economics while addressing challenges commanding government budgets—an attractive combination for technology companies accustomed to commercial markets.
Entry pathways differ in capital requirements and risk profiles. DIU programmes like HyCAT (Hypersonic and High-Cadence Airborne Testing) connect startups directly with Pentagon testing and validation, providing both technical feedback and procurement visibility. SBIR/STTR grants offer non-dilutive funding for early-stage development—typically $150K-2M for Phase I/II—reducing dilution while proving technical feasibility. For larger technology companies, strategic investment in defense startups provides exposure without building internal capabilities. New “as a service” contract structures compensate for outcomes versus products—for example, paying for satellite imagery coverage rather than purchasing satellites—reducing capital intensity while creating recurring revenue.
Consider three entry approaches and their requirements. Direct development requires security infrastructure, export control expertise, and cultural alignment with defense work. Strategic investment (taking equity positions in defense startups) requires portfolio construction skills and patience for longer exit timelines. Technology licensing or partnership requires negotiation capabilities and clear IP boundaries. Each pathway suits different organisational contexts—evaluate capabilities, risk tolerance, and strategic fit before committing.
Frameworks: Investment analysis covers build-buy-partner decision frameworks and strategic investor approaches.
Defense innovation introduces risks including IP theft, insider threats, export violations, reputational concerns, and government contract dependence. The L3Harris case—where executive Peter Williams sold eight zero-day exploits to Russian brokers, causing $35 million in damages—illustrates insider threat severity. Additional risks include long procurement timelines (averaging 5-7 years from prototype to production), political and budget uncertainty that can eliminate programmes mid-development, security clearance requirements creating operational constraints, and ethical considerations around military applications. Effective risk management requires technical controls, governance frameworks, and cultural approaches that balance security with organisational trust.
These risks fall into three categories requiring different mitigation approaches. First, supplier capability risks: can new entrants meet defense needs at affordable cost under harsh operational conditions? New defense solutions must incorporate step-change improvements in affordability while delivering advanced capabilities under weight, power, and environmental constraints that exceed commercial requirements. Second, demand signal risks: will customers procure at scale? The DoD comprises hundreds of stakeholders across services, commands, and acquisition offices—each with different pain points, risk tolerances, and procurement approaches. Converting pilot programmes to production contracts requires navigating this complexity. Third, regulatory and security risks: export controls (ITAR/EAR), security clearances, and supply chain transparency create operational overhead absent in commercial technology businesses.
Cross-border regulations add complexity despite recent coordination improvements. AUKUS enhances allied coordination on technology sharing, but jurisdictional differences in how various nations set and apply restrictions still complicate compliance. Technology companies must clearly define and continuously monitor their network of third- and fourth-party suppliers, understanding ultimate destination and end-user of goods and services. Exit environment compounds these risks: without clear pathways to program of record or strategic acquisition, valuations rely on optimistic projections that may not materialise.
Security deep-dive: Insider threat lessons and protecting trade secrets provides complete analysis and insider threat programme implementation guidance.
Peter Williams, L3Harris Trenchant General Manager, sold eight zero-day exploits to Russian broker Operation Zero over three years before detection—a $35 million breach highlighting gaps in monitoring privileged access and implementing insider threat programmes. Williams headed the division developing exploits exclusively for US and Five Eyes governments while simultaneously overseeing the internal leak investigation—a conflict revealing detection challenges from trusted insiders with authorised access. He received $1.3 million in cryptocurrency from Operation Zero (“the only official Russian zero-day purchase platform”) before federal investigators identified the breach. The case demonstrates that technical monitoring, behavioural analytics, and security culture are essential, not optional, for organisations handling sensitive technology.
Zero-day exploits—software vulnerabilities unknown to vendors or the public—command high prices because they provide temporary asymmetric advantages. Attackers can penetrate systems while defenders remain unaware of the vulnerability, unable to develop patches or countermeasures. In the L3Harris case, the exploits Williams sold enabled surveillance and offensive cyber operations against targets that presumed their systems secure. Federal prosecutors characterised Operation Zero as part of “the next wave of international arms dealers,” reselling exploits to non-NATO buyers including Russian government entities.
As offensive cyber capabilities become more valuable and contested, insider threat programmes play increasingly critical roles. The Williams case functions as a warning rather than an anomaly. Technical safeguards like privileged access management and data loss prevention are necessary but insufficient without robust human-centric security measures: behavioural analytics identifying unusual patterns, security clearance investigations and periodic renewals, clear reporting mechanisms for suspicious activity, and organisational culture that balances security monitoring with trust. These lessons apply beyond defense contractors—any organisation with valuable intellectual property faces similar risks from insiders with authorised access and financial motivation.
Complete analysis: L3Harris case study and insider threat lessons covers detection failures, legal consequences, warning signs, and practical implementation guidance for insider threat programmes.
Evaluate defense opportunities through structured frameworks examining strategic fit, technical feasibility, market opportunity, regulatory complexity, and organisational readiness. Key questions: Does your technology have genuine defense applications beyond superficial dual-use claims? Can you navigate procurement timelines measured in years and security requirements including clearances and export controls? Do you have risk tolerance for implementing insider threat programmes and accepting government contract dependence? Is defense work consistent with organisational culture and stakeholder values? The answers determine whether to pursue, partner, or pass on specific opportunities.
Start with decision clarity about who decides defense pursuit and who provides input. A RACI matrix (Responsible, Accountable, Consulted, Informed) formalises roles and prevents future conflicts about authority and process. CTOs should guide organisations through the strategic evaluation using frameworks that assess technology stack investments against scalability economics, developer productivity, innovation enablement, technical risk profile, and ecosystem advantages. Defense capabilities can be built internally, acquired through M&A, or accessed via partnership—each pathway suits different strategic contexts.
Success in defense requires understanding how government buys, who the decision-makers are, and how to align with long-term programmes of record. Defense operates differently from commercial markets—having superior technology gets you halfway to a contract, but the other half requires specialised domain knowledge about qualification processes, stakeholder management across service branches, and patience for procurement timelines. Understanding paths to acquisition, timing budget windows, and identifying the right customer within the DoD’s hundreds of stakeholders requires expertise that most technology companies lack initially.
Organisational readiness equals technical capability in importance. Do you have security infrastructure for handling classified information? Export control expertise for ITAR/EAR compliance? Cultural alignment with mission-driven work and acceptance of ethical considerations around military applications? Defense offers sticky revenue streams, mission-driven talent attraction, and strategic partnership opportunities—but introduces insider threat requirements, export compliance complexity, and budget dependence on political cycles. Honest appetite assessment prevents costly false starts when reality diverges from initial expectations.
Resources: Technology deep-dive on hydrogen propulsion assesses engineering foundations, startup blueprint from Hypersonix shows the pathway from research to Pentagon, investment analysis and funding landscape covers decision frameworks, and security lessons from the L3Harris case address risk management.
Defense innovation through 2030 will be shaped by ongoing Great Power Competition between the US, China, and Russia, autonomous systems proliferation across all military domains, hypersonic weapons development and counter-hypersonic defenses, and continued expansion of government-private partnership models. NATO’s 32 members pledged 5% GDP defense spending by 2035 (up from 2-3% currently), with portions explicitly flowing to startup innovation programmes. Expect expanded AUKUS technology sharing beyond submarines to include hypersonics and counter-hypersonic systems, more mega-rounds for companies demonstrating capability delivery, and technology focus on autonomy, electronic warfare, quantum applications, and cyber capabilities. Insider threat awareness and security requirements will intensify following high-profile breaches like L3Harris.
Rising geopolitical tensions and battlefield adaptations during the Ukraine conflict highlight the urgency of technological advancement. Ukraine demonstrated how commercial drones, Starlink communications, and open-source intelligence fundamentally change modern warfare—lessons that accelerate military adoption of autonomous and AI-driven technologies. Militaries worldwide are shifting toward distributed, technology-intensive force structures and away from platform-centric warfare built around expensive, vulnerable assets like aircraft carriers. China’s hypersonic glide vehicle advances and Russia’s Avangard and Kinzhal operational systems drive Western development urgency. As Pentagon DIU programme manager stated about HyCAT: “Right now, we test hypersonic systems once a year. We need to be testing them weekly.”
Investment trends reflect this strategic shift. Government co-investment models will expand as more nations establish sovereign investment vehicles modelled on NRFC and NATO Innovation Fund. Mega-rounds will flow to companies demonstrating not just prototypes but operational capability and pathway to programme of record. Exit markets will either mature through strategic acquisitions and IPOs, or valuations will correct downward if the gap between prototype and production proves too wide. The Australian context: sovereign capability development positions Brisbane as a regional defense hub, leveraging research institutions, government co-investment, and AUKUS market access. Regulatory evolution will tighten export controls and security clearance requirements as insider threat awareness increases across allied governments.
Future insights: Hypersonic technology roadmap and hydrogen propulsion explained explores next-generation capabilities, Brisbane ecosystem emergence and the Australian hypersonic startup provides a replicable model, investment trend analysis and funding landscape examines capital flows and exit pathways, and evolving threat landscape and protecting trade secrets addresses security imperatives.
How Hydrogen-Powered Scramjets Are Enabling Mach 12 Flight
Deep technical explainer covering scramjet physics, hydrogen propulsion advantages, thermal management at temperatures exceeding 1,800°C, 3D printing applications, and material science challenges. If you need to understand the engineering fundamentals enabling hypersonic breakthroughs, start here.
Best for: Readers seeking engineering fundamentals and technical feasibility assessment Content Type: Conceptual/Explainer Reading Time: 10-12 minutes
Hypersonix Launch Systems – How an Australian Startup Is Building Hypersonic Aircraft with NASA and the Pentagon
Comprehensive case study of Hypersonix’s journey from University of Queensland research to $46 million Series A and Pentagon HyCAT testing programme. Covers founder background, technology differentiation (SPARTAN engine), investor composition (NRFC, High Tor Capital, Saab), product roadmap (DART → VISR → DELTA VELOS), and lessons for deep tech entrepreneurs.
Best for: Readers evaluating startup pathways and government-private partnerships Content Type: Case Study Reading Time: 12-15 minutes
Defense Tech Investment in 2025 – Where Government and Venture Capital Are Backing Breakthrough Innovation
Analysis of defense tech funding landscape covering government co-investment models (NRFC, Office of Strategic Capital, NATO Innovation Fund), investor types (national security VCs, strategic corporates, sovereign funds), Valley of Death solutions, build-buy-partner frameworks, and strategic partnership approaches.
Best for: Readers making investment or partnership decisions Content Type: Analysis/Guide Reading Time: 9-11 minutes
The L3Harris Insider Threat Case – What the Peter Williams Guilty Plea Reveals About Protecting Trade Secrets
Detailed examination of the Peter Williams insider threat case (eight zero-day exploits sold to Russian Operation Zero, $35 million damages), detection failures, legal consequences, warning signs, and practical guidance for building insider threat programmes. Balances security requirements with organisational trust.
Best for: Readers concerned with security and risk management Content Type: Case Study + Implementation Guide Reading Time: 11-14 minutes
Hypersonic technology refers to flight systems operating above Mach 5 (five times the speed of sound), enabling rapid response capabilities and evasion of traditional defense systems. Unlike supersonic flight (Mach 1-5), hypersonic systems sustain extreme speeds through scramjet engines enabling supersonic combustion. For defense applications, this translates to strategic advantages in surveillance, reconnaissance, and response times.
Learn more: How Hydrogen-Powered Scramjets Are Enabling Mach 12 Flight
Yes, particularly in emerging technology domains where agility and innovation speed provide advantages. Startups demonstrate this pathway through breakthrough technology differentiation, strategic government programme access (like Pentagon HyCAT), and mixed government-private funding that bridges the Valley of Death. Success requires clear technology advantages over incumbent solutions, patient capital comfortable with 5-7 year development timelines, and domain expertise navigating defense procurement processes.
Learn more: Hypersonix Launch Systems case study
The NRFC is Australia’s government co-investment fund providing equity investments in strategic industries including defense, advanced manufacturing, and renewable energy. Its $10 million investment in Hypersonix’s Series A marked the fund’s first defense sector commitment, partnering with international VCs and strategic corporates. This model mirrors the US Office of Strategic Capital and NATO Innovation Fund approaches: de-risking private investment while accelerating sovereign capability development.
Learn more: Defense Tech Investment in 2025
A zero-day exploit is a software vulnerability unknown to the vendor or public, giving attackers temporary asymmetric advantage before defenses can be developed. These vulnerabilities command high prices because they enable system penetration while defenders remain unaware, unable to patch or implement countermeasures.
Learn more: The L3Harris Insider Threat Case
AUKUS (Australia-United Kingdom-United States trilateral partnership) enables technology transfer, joint development programmes, and coordinated procurement in priority areas including hypersonics and counter-hypersonic systems. For Australian companies, AUKUS creates pathways to US and UK markets totalling over $1 trillion in annual defense spending, Pentagon testing programmes, and allied procurement budgets. It amplifies Australia’s strategic relevance well beyond its $50 billion annual defense budget.
Coverage across articles: Mentioned throughout this overview, detailed in Hypersonix case study and investment analysis.
Yes—any organisation handling valuable intellectual property faces insider threat risks, not just defense contractors. The L3Harris case involved a General Manager with authorised access, not an external hacker, highlighting that trusted insiders pose significant risks when internal controls fail. Effective programmes combine technical controls (monitoring privileged access, detecting anomalous behaviour), governance frameworks (security clearances, access policies), and cultural approaches (reporting mechanisms, awareness training).
Learn more: The L3Harris Insider Threat Case
The Defense Innovation Unit is a US Department of Defense organization accelerating commercial technology adoption through streamlined procurement processes, prototype funding, and Other Transaction Agreements (OTAs). DIU manages programmes like HyCAT (Hypersonic and High-Cadence Airborne Testing) that connect startups with Pentagon testing and validation pathways. It represents DoD’s effort to access startup innovation without traditional procurement barriers.
Coverage: Mentioned throughout, detailed in Hypersonix case study and investment analysis
Use a structured framework examining: (1) Strategic fit—does your technology have genuine defense applications? (2) Technical feasibility—can you meet performance and security requirements? (3) Organisational readiness—do you have security infrastructure and regulatory expertise? (4) Risk tolerance—can you navigate export controls, procurement timelines, and insider threat requirements? (5) Values alignment—is defense work consistent with organisational culture and stakeholder expectations?
Framework details: Covered in the “How Should Technology Leaders Approach Defense Tech Opportunities?” section above and throughout investment analysis
Defense innovation offers opportunities for technology companies that understand its complexities. Breakthrough technologies, government-private partnerships, and geopolitical imperatives create space for agile innovators in emerging domains. But opportunity without strategic clarity leads to misallocated resources and unrealised objectives.
The frameworks, case studies, and analyses linked throughout provide decision-support tools for evaluating alignment with your capabilities, risk tolerance, and values. Your starting point depends on your immediate questions:
If evaluating technology feasibility and engineering challenges → Start with How Hydrogen-Powered Scramjets Are Enabling Mach 12 Flight to understand the technical foundations and complexity levels
If assessing market entry strategy and startup pathways → Start with Hypersonix Launch Systems case study to see how research transitions to commercial capability
If making investment or partnership decisions → Start with Defense Tech Investment in 2025 to understand investor motivations, co-investment models, and exit pathways
If concerned about security and compliance requirements → Start with The L3Harris Insider Threat Case to understand risk management imperatives
The landscape will continue shifting through 2030 as government commitments increase, technologies mature, and security requirements tighten. Whether engaging directly, investing strategically, or monitoring for competitive intelligence, understanding this ecosystem provides advantage. Each cluster article provides depth for specific decision contexts. Return here to explore adjacent topics or reassess strategic fit as your understanding evolves.