In 2018 Google walked away from a Pentagon AI contract after more than 4,000 employees protested that the company “should not be in the business of war.” Then, in the first five months of 2026, defence tech startups raised more than $14.6 billion, beating 2025’s full-year record before summer. Defence tech now competes with your business for capital, talent and government-adjacent customers, so mistaking its economics for SaaS is the expensive misread. Here is the three-part descent: the reversal, the inverted economics, and the gap that kills companies after they win military contracts. For the wider picture, the full defence tech boom overview is the place to start.
What turned defence tech from a Silicon Valley taboo into a venture capital asset class?
Defence tech turned from taboo into a venture capital asset class after 2022, when the Ukraine war proved that a weapon’s value now lives in its software rather than its metal, and the US Department of Defense opened procurement to startups. That shift made defence revenue visible enough to underwrite, driving more than $14.6 billion raised globally in the first five months of 2026.
Project Maven: the moment defence became taboo
Google’s Project Maven contract applied AI to drone imagery, and the backlash forced the company to let it lapse; Anduril, founded a year earlier, was soon headlined as the most controversial startup in tech. For years after, investors saw defence as too slow, bureaucratic and ethically fraught, leaving startups largely shut out of a market owned by legacy primes.
Why 2022 changed the calculus
Russia’s invasion of Ukraine put cheap drones and AI-enabled systems onto the battlefield, and CB Insights’ Jason Saltzman says Ukraine “demonstrated drone and autonomous system effectiveness in real combat”. The Pentagon did the rest. Through the Defense Innovation Unit and other transaction authorities, it opened a path for startups to win contracts in months instead of years, outside the slow acquisition rules built for the legacy primes. Palantir settled the exit question by riding government contracts to the S&P 500, while US-based startups drew roughly $38 billion in the first half of 2025 alone and specialist funds rounded out the funding lifecycle. NEA’s Aaron Jacobson calls it compounding confidence: Palantir and Anduril succeeded, and investors now believe more innovation is possible. The money is real. The open question is whether the economics behind it behave like anything venture investors already know.
How do defence tech economics differ from SaaS or consumer-software economics?
Defence tech economics invert SaaS. Instead of many customers paying recurring subscriptions at 80%-plus software margins, a defence startup sells multi-year contracts to one budget-backed buyer, absorbing hardware capital costs and long sales cycles in exchange for sticky, contracted forward revenue and durable moats.
Contract revenue vs subscription revenue
SaaS revenue arrives monthly and churns; defence revenue arrives as multi-year budget commitments. The buyer is a single dominant customer (a monopsony), the US Department of Defense, which recently spent roughly $997 billion on the military in a single year. That makes revenue forecastable but concentrated, moving through congressional appropriations rather than a sales pipeline.
Legacy primes bill cost-plus at a fixed margin around 9%, “the inverse of the quintessential VC-backed startup”, which invests its own capital up front and earns high software margins later. Hardware companies carry factories and manufacturing capex; software-only players carry far less.
Why investors accept hardware-heavy economics
Overmatch Ventures’ Morgan Hitzig calls defence tech “in many ways the inverse of AI or SaaS economics”: heavy capex up front, but strong sales efficiency once deployed and proven. Bessemer’s Janelle Teng Wade counters that sales cycles are long, lumpy and hard to forecast, bending with political funding cycles.
What investors buy is a moat. Once a company breaks into a Program of Record, a fully funded line item in the DoD budget, contracts are large and enduring, and the bureaucracy that blocked entry starts blocking competitors. Non-dilutive R&D money softens capital cost without diluting equity, so valuations run on contracted forward revenue and run-rate rather than SaaS churn. Concentrated government relationships affect the risk attached to future revenue. Whether that pricing holds is the question behind the revenue multiples those economics now support. Before any of that pricing matters, a won contract still has to become production revenue, and that conversion has its own name.
What is the “valley of death” in defence tech, and why do military contracts not save you from it?
The valley of death is the funding gap between winning a prototype or pilot contract and reaching production-scale revenue. Contracts do not save companies because cancellations, appropriation delays and bridge-financing gaps can leave startups unfunded for years before a Program of Record. Shield AI is crossing it through hybrid hardware-plus-software revenue.
Why a contract does not close the valley
A startup can win a pilot or R&D grant and still sit unfunded for years, because the budget never aligns behind the technology. Good technology dies when it never secures a formally funded programme, even if it works. Continuing Resolutions have hit every budget cycle for five years, keeping the gap open. The milestone that closes it is a Program of Record.
How Shield AI crossed it
Shield AI is the case to study. It pairs V-BAT uncrewed aircraft with Hivemind autonomy software across 30 vehicle classes, doubled orders and revenue year over year in FY26 and announced $2 billion in funding at a $12.7 billion valuation. The hybrid structure matters: recurring software revenue carries the company across the hardware valley, while pure-software Palantir carries less capital intensity but a thinner moat. Shield AI remains pre-Program of Record, so this is a crossing in progress. Asset-class status has repriced that risk; the gap itself remains.
So where does that leave you? Defence tech became investable when investors learned to price a different, inverted set of risks. The taboo broke because a geopolitical shock and a buyer that made defence revenue visible arrived together. The economics run on forward revenue and run-rate, and the Program of Record separates a real business from a headline round. The question that now matters is which companies can cross the valley to production revenue. Read the rest in the wider cluster on defence technology.
Frequently Asked Questions
What does software-defined warfare actually mean?
Software-defined warfare means the value of a weapon system lives in its software, not its metal. A drone or armoured vehicle becomes a platform that can be upgraded, re-tasked and given new behaviours through code updates, the way a phone gains new apps. Ukraine proved the point: cheap, rapidly iterated software drones repeatedly outmatched expensive hardware that could not adapt between battles.
Is defence tech mostly about weapons, or does it include software and AI?
Defence tech spans far more than weapons. The label covers artificial intelligence, autonomy, sensors, cyber security, space, electronic warfare and the advanced manufacturing behind them. Anduril and Shield AI, two of the sector’s best-known names, sell systems whose core is software with hardware wrapped around it. Investors apply the label broadly, so very different companies now compete for the same procurement budgets and the same venture dollars.
Was ethics the only reason venture capital avoided defence tech for so long?
Ethics was only part of it. The Project Maven-era stigma was real, but structural problems mattered as much: almost no defence startups reached public markets before Palantir’s 2020 listing, so exits were scarce; timelines outran the typical ten-year fund; and the buyer was a single, unpredictable government. The post-2022 changes fixed the structure, not the morality. That is why capital flooded in while some investors still refuse weapons work on principle.
Is it true that big tech companies still refuse military work after Project Maven?
Not industry-wide. Google withdrew from Project Maven in 2018 and wrote weapons out of its AI principles, but its rivals did not follow. Microsoft and Amazon continued pursuing major defence contracts, and many of the startups that now define the sector were founded specifically to serve the Pentagon. The ‘never again’ moment turned out to be Google’s decision rather than Silicon Valley’s, and the post-2022 reversal proved it.
What is an Other Transaction Authority and why does it matter for defence startups?
An Other Transaction Authority, or OTA, lets the Department of Defense fund startups without the slow acquisition rules designed for giant prime contractors. Prototype OTAs can be awarded in months rather than years, startups often keep their intellectual property, and the money does not dilute equity. This mechanism, championed by the Defense Innovation Unit, is a big part of why defence revenue became visible and forecastable enough for venture capital to underwrite.
What is dual-use technology and why do defence investors like it?
Dual-use technology works for military and civilian customers at once: autonomy software that guides both drones and delivery vehicles, or AI used for battlefield analysis and medical imaging. Investors value it because a commercial revenue stream softens dependence on a single government buyer and keeps the business closer to familiar market economics. Palantir is the archetype, built on government contracts but pointing to commercial revenue as its long-run growth engine.
How is a defence tech startup different from a traditional contractor such as Lockheed Martin?
The primes, companies such as Lockheed Martin and Raytheon, build enormous platforms like fighter jets and missile systems under cost-plus contracts that stretch for decades. Defence tech startups build smaller, software-led systems, iterate in months and win prototype work at a speed and price the primes cannot match. The two worlds now collide, with primes partnering with, investing in and sometimes acquiring the startups that threaten them.
Is the defence tech boom happening outside the United States?
Mostly American in scale, global in effect. The United States dominates because its Department of Defense is the largest buyer and has done the most to open procurement to young companies. Europe has lifted military spending sharply since 2022, and allies such as Australia are running their own procurement reforms, but the venture money still concentrates in US startups. Expect the pattern to spread as allied budgets and startup ecosystems catch up.
Can ordinary investors buy into defence tech, or is it reserved for venture funds?
Most defence tech companies are private, so the flagship names stay with venture funds, crossover investors and accredited individuals. Ordinary investors can still gain exposure through listed companies such as Palantir, through specialist defence funds and ETFs, or by watching for future public listings. Direct startup investing generally demands both capital and accreditation, which is why the asset-class conversation has remained largely institutional so far.
Has the engineering talent taboo on defence work really disappeared?
The stigma has faded but not vanished. Since 2022, far more engineers are willing to work on military technology, drawn by mission, funding and the chance to build autonomy and AI at scale. But some veterans of the Project Maven era still refuse weapons work, and the argument has partly moved rather than ended, resurfacing around lethal autonomy and export controls instead of blanket refusals.
