What the Nuclear Power Renaissance Means for the Future of AI Infrastructure

Global data centre electricity consumption surpassed 1,000 TWh in 2025 — more than Japan’s entire national grid. The forcing function is AI: training and inference workloads require continuous, 24/7 firm power that solar and wind alone cannot reliably supply at scale. The result is the largest nuclear contracting wave since the Cold War — Microsoft, Google, Meta, Amazon, and Oracle have committed to more than 10 GW of nuclear capacity through a combination of long-term power purchase agreements and SMR deals.

Whether or not your organisation owns a server, these decisions are already shaping cloud costs, AI vendor options, and infrastructure risk across the market. This page maps the story. Each section below links to the deep-dive article on that dimension.

In this series:

• Why Tech Giants Are Going Nuclear
• Inside the SMR Race
• The Broken Power Grid
• The $98 Billion Hidden Constraint
• Nuclear Power Policy in the Age of AI
• Nuclear vs Everything Else
• AI Power Risk Is Now a Board-Level Issue

Why are tech companies suddenly investing in nuclear power?

AI workloads require continuous, always-on power that solar and wind cannot reliably deliver at the scale modern data centres demand. Nuclear is the only carbon-free generation source that operates at gigawatt scale with the reliability these facilities require. When the public grid cannot deliver new capacity fast enough — interconnection queues now run seven to thirteen years in key markets — the largest technology companies have begun contracting directly with nuclear operators and SMR developers to secure power outside the grid entirely.

Full analysis: Why Tech Giants Are Going Nuclear

Which technology companies have signed nuclear deals — and for how much?

Microsoft restarted Three Mile Island Unit 1 under a 20-year contract for 835 MW. Meta announced 6.6 GW across three deals in January 2026 — covering Vistra‘s Ohio plants, TerraPower SMRs, and Oklo microreactors. Google contracted 500 MW from Kairos Power across six to seven molten salt reactors; Amazon committed to X-energy SMRs via Energy Northwest in Washington; and Oracle announced plans for a 1 GW campus powered by three SMRs.

Deal breakdown and analysis: Why Tech Giants Are Going Nuclear

What is a Small Modular Reactor (SMR) and when will one actually be available?

A Small Modular Reactor is a nuclear plant designed to generate under 300 MW per unit — roughly a third the output of a conventional reactor — with modules factory-built and assembled on site rather than constructed from scratch. No commercial SMR is yet operating in the United States; the most credible deployment timelines for the hyperscaler-backed designs cluster between 2030 and 2035. Existing nuclear plant restarts are the only nuclear capacity that can be online before 2028. The two most significant near-term risks to those timelines are the HALEU fuel supply chain — a specialised uranium enrichment product with limited domestic production — and NRC licensing processes that remain untested for advanced reactor designs at commercial scale.

Technology comparison: Inside the SMR Race

Why can’t the US power grid just supply what data centres need?

The US grid was not built for the density or pace of data centre load growth that AI is now generating. Seven of thirteen major grid regions are projected to operate below safety margins by 2030, and the queue to connect new generation or large loads to the transmission network now averages seven to ten years in established markets. This is the structural reason hyperscalers are pursuing behind-the-metre power — building their own generating capacity adjacent to data centre campuses to bypass the grid entirely.

Infrastructure deep dive: The Broken Power Grid

Why can’t renewables alone power AI data centres?

Wind and solar are intermittent — they generate electricity only when the wind blows or the sun shines — and cannot provide the continuous baseload power AI inference workloads require. Battery storage at the scale needed to bridge renewable intermittency for a large data centre campus does not yet exist at an economically viable cost. Until large-scale storage matures, nuclear and natural gas are the only firm-power options available; nuclear’s carbon-free profile makes it the preferred long-term answer for hyperscalers with sustainability commitments.

Full comparison: Nuclear vs Everything Else

What are the hidden constraints slowing AI infrastructure buildout beyond power generation?

Beyond grid capacity, two physical constraints are quietly throttling AI infrastructure expansion. Grid interconnection queues mean that even if new nuclear or renewable capacity is built, connecting it to the transmission network can take a decade. Separately, $98 billion in planned data centre projects were blocked by local zoning and community opposition in Q2 2025 alone — a constraint that most technology leaders have not yet factored into infrastructure planning.

• Grid bottleneck: The Broken Power Grid
• Zoning and opposition: The $98 Billion Hidden Constraint

What are the safety concerns with fast-tracking nuclear development for AI?

The AI Now Institute‘s 2025 report “Fission for Algorithms” documents specific regulatory shortcuts being proposed or implemented to accelerate SMR approvals — including a new DOE pathway that authorises reactors without independent NRC review, and executive orders targeting foundational radiation safety models. Proponents argue that advanced SMR designs are inherently safer than legacy reactors and that streamlined rules are proportionate to lower risk profiles. Critics argue that speed-driven licensing erodes the safety culture and independent oversight that have kept civilian nuclear power safe for decades.

Regulatory and policy context: Nuclear Power Policy in the Age of AI

How are governments in the US and UK responding to AI’s energy demands?

The US response has combined regulatory acceleration — the ADVANCE Act, Genesis Mission, DOE loan guarantees — with emergency grid measures; the UK launched its Advanced Nuclear Framework in February 2026, designated Wylfa as the first new nuclear site, and established Great British Energy-Nuclear with backing from the National Wealth Fund. Both governments frame nuclear as essential infrastructure for AI leadership. The policy race has direct implications for organisations with operations on both sides of the Atlantic.

US and UK policy deep dive: Nuclear Power Policy in the Age of AI

What does this mean for companies that do not own a data centre?

If your company uses cloud infrastructure to run AI workloads — which describes the overwhelming majority of organisations — the nuclear-AI story is already affecting your cost and risk profile. The capital expenditure hyperscalers are committing to nuclear power, combined with grid and zoning constraints limiting data centre supply, flows directly into cloud pricing and regional availability. Mark Zuckerberg’s 2025 statement that “power will be the bottleneck” was not a comment about Meta’s internal planning — it is a preview of the constraint that will shape AI infrastructure access across the market.

Strategic framework: AI Power Risk Is Now a Board-Level Issue

Resource Hub: Nuclear Power’s AI Renaissance Library

The Forces Driving Nuclear’s Return

• Why Tech Giants Are Going Nuclear — The AI power crisis, the hyperscaler deals, and why renewables alone cannot meet baseload demand. Read the full analysis
• Inside the SMR Race — A technology comparison of the four leading small modular reactor designs, with cost and timeline data translated for non-specialists. Read the comparison
• Nuclear vs Everything Else — How nuclear stacks up against solar, wind, geothermal, and gas on timeline, cost per MWh, reliability, and carbon profile. Read the comparison

The Constraints Shaping the Buildout

• The Broken Power Grid — Why grid interconnection queues are forcing hyperscalers toward behind-the-metre power — and what that means for cloud region resilience. Read the infrastructure analysis
• The $98 Billion Hidden Constraint — The zoning and community opposition story that most technology leaders are not factoring into infrastructure planning. Read the counter-narrative
• Nuclear Power Policy in the Age of AI — How the US and UK are racing to enable the nuclear buildout — and what the regulatory shortcuts mean for safety. Read the policy analysis

Strategy and Decision Support

• AI Power Risk Is Now a Board-Level Issue — A framework for assessing AI infrastructure risk, evaluating cloud provider energy positions, and framing power risk for your CEO and board. Read the strategic framework

Frequently Asked Questions

Is nuclear power actually making a comeback because of AI?

Yes — and the scale of recent commitments suggests it is not temporary. More than 10 GW of nuclear capacity has been contracted by US hyperscalers since 2023, across both restarted existing plants and long-term SMR agreements. The International Energy Agency‘s April 2026 data confirms that data centre electricity consumption surpassed 1,000 TWh globally in 2025, with AI workloads as the primary growth driver.

What is a power purchase agreement (PPA) and how do tech companies use them for nuclear?

A power purchase agreement is a long-term contract — typically 15 to 25 years — in which a buyer commits to purchasing a set quantity of electricity from a specific generator at an agreed price. Tech companies use PPAs to make nuclear projects financially viable: by guaranteeing a revenue stream for 20+ years, they give nuclear operators and SMR developers the certainty required to finance construction. The major hyperscaler deals — Microsoft’s reactor restart contract, Meta’s Vistra agreement, and Google’s Kairos arrangement — are all structured as long-term PPAs.

What does “behind-the-metre” power mean for data centres?

Behind-the-metre generation means building a power source directly adjacent to a facility so that electricity never flows through the public grid. For data centres, this model bypasses the years-long interconnection queue entirely — the operator effectively becomes its own utility. The tradeoff is substantial upfront capital and the regulatory complexity of siting a power plant alongside a data centre campus.

How does AI energy demand in the US compare to China’s?

The US leads in hyperscaler-driven nuclear and clean energy contracting; China is responding with scale and speed across multiple generation types simultaneously. China is adding more than 400 GW of new power capacity annually, deploying both the world’s largest renewables build-out and new coal plants, while also pursuing its own advanced reactor programme. The competition for AI infrastructure energy is effectively geopolitical, with both nations treating data centre power as a strategic resource.

Will AI data centres cause electricity prices to go up for everyone else?

The evidence suggests some upward pressure is likely in regions with the highest data centre density, but the effect is not uniform. In heavily saturated markets like Northern Virginia, grid stress from data centre load is already contributing to higher wholesale electricity prices and reliability concerns. Regions with excess generating capacity are less exposed. The policy debate around equitable cost allocation — whether hyperscalers should bear more of the grid upgrade cost — is active but unresolved.

When will SMRs actually be available to buy power from?

The most realistic commercial SMR timelines for the hyperscaler-backed US designs cluster around 2030–2035. Kairos Power’s first unit targeting 2030, TerraPower’s Natrium plant in Wyoming targeting 2030+, and X-energy’s Energy Northwest project targeting the early 2030s represent the current credible pipeline. No commercial SMR is operating in the United States today; NuScale is the only design with full NRC certification, and its original deployment project was cancelled in 2023 due to cost escalation.