Meta’s Nuclear Bet: How Firm Power Became Core to AI Infrastructure

TL;DR: Meta signed 20‑year power purchase and investment agreements with Vistra, Oklo, and TerraPower to back its Prometheus AI data center and regional AI campuses with predominantly nuclear-generated electricity. The deals — supporting up to about 6.6 GW by 2035 — show hyperscalers treating firm clean energy as strategic infrastructure, not just an ESG add‑on.

Quick facts executives need to scan

• Deal length: 20‑year corporate PPAs and investments.
• Scale: Up to ~6.6 GW by 2035 (equivalent cited: ~5 million homes).
• Vistra: ~2,609 MW between 2026–2034 (2,176 MW existing + ~433 MW added); >2.1 GW sourced from two Ohio plants.
• Oklo: Supports a 1.2 GW modular reactor campus in Pike County, Ohio for regional AI hubs and the New Albany AI supercluster.
• TerraPower: Funding for two Natrium units (~690 MW by 2032) and energy rights for up to six more (potentially ~2.1 GW by 2035).
• Market moves: Vistra stock rose ~10.5% (to $166.38) with trading volume nearly doubling; Oklo rose ~7.9% (to $105.32).

Jargon decoded (one-line definitions)

• PPA (power purchase agreement) — a long-term contract to buy electricity that guarantees revenue to generators so they can get financing.
• Firm power — electricity that’s continuously available on demand (as opposed to intermittent wind/solar).
• Uprate — increasing output of an existing reactor through engineering upgrades.
• Baseload — steady, predictable power supply; nuclear typically provides high baseload availability.
• Natrium — TerraPower’s advanced reactor design; an example of next-gen nuclear that pairs reactor cores with molten-salt energy storage.
• SMR (small modular reactor) — factory-built, smaller nuclear units that can be deployed modularly.

Why nuclear power for AI infrastructure?

Hyperscale AI consumes large, steady streams of electricity. Models like Prometheus need continuous power for GPUs running 24/7 — not a pattern that maps neatly onto solar or wind without massive, costly storage. Nuclear delivers high capacity factors (steady output across time) and predictable availability, which makes it an attractive complement to renewables when you’re sizing facilities for sustained inference and training workloads.

Beyond reliability, there’s finance and procurement logic. Long-term PPAs and direct investments reduce revenue risk for plant operators and advanced‑reactor builders. That revenue certainty helps projects clear regulatory hurdles and attract capital — particularly important for advanced designs and uprates that otherwise struggle to secure financing.

What Meta actually signed (the deal mechanics)

The package mixes offtake contracts and direct investments across incumbent utilities and advanced reactor developers. Highlights:

• Vistra: Meta will buy roughly 2,609 MW across 2026–2034, blending existing plant output with capacity added via uprates and expansions. Meta expects more than 2.1 GW to come from Vistra’s two Ohio sites, plus contributions from Pennsylvania plants.
• Oklo: The deal supports a planned 1.2 GW modular reactor campus in Pike County, Ohio — intended to serve regional AI centers, including the New Albany AI supercluster.
• TerraPower: Meta will fund two Natrium units producing ~690 MW by 2032, and secure rights for up to six more Natrium modules, potentially reaching ~2.1 GW by 2035.

“This commitment from Meta provides Vistra the certainty needed to invest in these plants and communities and bring new nuclear generation online for the grid — through uprates at our existing plants.” — Jim Burke, CEO at Vistra

“[These deals] make [Meta] one of the largest corporate consumers of nuclear power in U.S. history.” — Joe Kaplan, Meta’s chief global relations officer

Meta’s head of global energy, Urvi Parekh, framed the transactions as securing the reliable power required to meet their AI ambitions — a sentiment that neatly captures the strategic shift: procurement teams are now infrastructure teams.

Market reaction and why investors care

Vistra’s share price jumped roughly 10.5% with trading volume nearly doubling. Oklo’s stock also rallied. The market response highlights an important dynamic: corporate PPAs can materially change project economics and investor sentiment overnight because they convert speculative projects into bankable assets.

For developers and utilities, that’s the point. Revenue‑backed projects are easier to finance and less risky to regulators. For hyperscalers, locking price and supply for decades hedges against volatile power markets and reputational risk tied to emissions targets.

Risks and unknowns executives should weigh

• Timing and delivery risk: Licensing, permitting, construction delays, and cost overruns can push timelines beyond the dates companies are counting on.
• Contract structure: How much power is physically earmarked for Meta versus how much is bought as grid-delivered energy will shape local prices and public perception.
• Transmission constraints: Power “flows through” regional grids, but interconnection bottlenecks or congestion can limit benefits to local communities or lead to curtailment.
• Political and regulatory hurdles: Advanced reactors face scrutiny and siting challenges that can slow rollout.
• Alternatives tradeoffs: Renewables plus storage are cheaper at scale for many loads, but for continuous, high-density AI loads the required storage becomes a steep capital and operational ask.

How this moves the needle for other hyperscalers and corporate buyers

Meta’s move signals a playbook: when AI scale demands predictable, continuous power, large buyers may combine long PPAs with equity-like investments in firm generation. Historically, hyperscalers leaned heavily on renewables PPAs to hit clean energy targets. This transaction mix shows a strategic pivot toward diversity — pairing renewables with firm, low‑carbon sources to manage risk and availability.

Expect procurement teams to consider a spectrum of levers: virtual PPAs for carbon accounting, physical offtakes for guaranteed delivery, equity stakes to accelerate vendor builds, and hybrid contracts that mix firm nuclear baseload with variable renewables plus storage. Each choice moves a different dial on cost, control, and reputational exposure.

Scenarios: three plausible outcomes

• Optimistic: Projects stay on schedule and bring low‑carbon, firm capacity online; regional reliability improves and energy price volatility declines, giving AI projects cheaper, stable power.
• Base case: Mix of uprates and new builds arrive with some delays and moderate cost overruns. Grid reliability benefits are regional and gradual; corporate buyers secure enough capacity for incremental AI growth.
• Pessimistic: Regulatory or construction setbacks push timelines beyond commitments, forcing buyers to rely on spot markets or backstop gas generation — eroding the strategy’s cost and emissions benefits.

Key questions and quick answers

• Who signed the deals?
  

  Meta signed 20‑year agreements with Vistra, Oklo, and TerraPower to supply its Prometheus AI data center and regional AI infrastructure with nuclear-generated electricity.

• How much capacity and by when?
  

  The package supports up to about 6.6 GW by 2035 — including ~2,609 MW from Vistra (2026–2034), a 1.2 GW Oklo campus, and TerraPower Natrium units (690 MW by 2032 and up to ~2.1 GW by 2035 with options).

• Why nuclear for AI?
  

  Nuclear provides firm, baseload clean power that complements intermittent renewables and meets continuous, high‑capacity electricity demands typical of AI superclusters.

• What are the main risks?
  

  Regulatory delays, permitting and construction risks, transmission constraints, and the specifics of contract structures (physical reservation vs. grid rights).

Actionable takeaways for boards and procurement leaders

• Start modeling firm power into TCO for AI: Run scenarios that combine nuclear PPAs, renewables, and storage to understand costs per kWh for sustained AI loads.
• Think beyond price: Evaluate contract types — physical offtake, virtual PPA, equity investment — for control, reputational impact, and financing leverage.
• Engage grid planners early: Transmission and interconnection are often the gating factors; coordinate with utilities and regulators before committing to site buildouts.
• Use long PPAs as leverage: Multi-decade offtakes can de-risk projects and speed permitting — but insist on contractual protections for delays and force majeure.
• Prepare communications for communities and regulators: Nuclear carries political sensitivity; transparent public engagement and local benefit commitments reduce opposition risk.

CFO takeaway: Locking firm, predictable energy via long PPAs can materially reduce volatility in AI operating expense and make capital planning for compute expansion more defensible.

Head of Infrastructure takeaway: Treat energy procurement as a strategic asset. Combining nuclear firm power with renewables and storage creates a resilient, low‑carbon stack for sustained AI workloads.

Meta’s commitments mark a strategic inflection: corporate energy procurement is moving from trading contracts to underwriting physical infrastructure. For companies building AI capacity, the question is no longer just “how green are we?” but “how reliably and affordably can we power the compute that runs our business?” Boards that answer that question proactively will shape where and how the next generation of AI systems gets built.