Meta's Nuclear Bet: Securing the AI Power S-Curve

The AI revolution is not just a software shift; it is a fundamental demand for physical infrastructure. At its core is an exponential growth curve for power. According to the Electric Power Research Institute, AI power demand in the US could increase tenfold by 2030, surging from a current 5 GW to more than 50 GW. This isn't a linear climb but a steep S-curve, driven by the massive energy needs of training frontier AI models. The scale is staggering: a single large training run could soon rival the output of a major power plant, demanding new approaches to grid planning and permitting.

Meta is positioning itself at the inflection point of this curve. The company's $600 billion digital infrastructure pledge by 2028 is a first-mover bet on securing the fundamental energy rails for this paradigm shift. Data centers are the physical manifestation of this bet, and the company is committing to build industry-leading AI data centers right here in the US. This massive capital outlay, expected to see a significant jump in 2026, is a direct play on the decade-long trajectory of AI power consumption.

The Prometheus supercluster, with its expected launch in 2026, is a key project driving this demand. To power this system, Meta has made landmark deals with three nuclear providers, securing a combined 6.6 GW of power by 2035. This isn't just about sourcing electricity; it's about locking in low-carbon capacity at scale to support the next generation of compute. The company is essentially building its own energy supply chain, recognizing that the AI race will be won by those who control the infrastructure that runs it.

The Infrastructure Bet: Building the Energy Rails

Meta's nuclear deals are a masterclass in infrastructure positioning. This is not a simple power purchase; it is a multi-pronged investment to build the fundamental energy rails for the AI S-curve. The company has secured a total of 6.6 gigawatts of power by 2035, a capacity equivalent to about six large nuclear reactors. This massive commitment is structured across three distinct project types, creating a diversified energy supply chain.

The core of the deal is a long-term, low-carbon anchor. Meta has signed 20-year power purchase agreements with Vistra to extend and expand three existing nuclear plants in the Midwest. This provides immediate, reliable capacity to power the Prometheus supercluster's launch in 2026. At the same time, the company is making strategic prepayments to accelerate the future. Its agreement with Oklo is a prime example, providing funding to advance Phase 1 of a 1.2 GW power campus in southern Ohio. This prepayment model de-risks advanced reactor development, a critical step for scaling next-generation technology.

The portfolio is a deliberate mix of proven and pioneering. The Vistra plants represent the existing, licensed fleet-essential for near-term power. The Oklo project targets advanced small modular reactors (SMRs), aiming for a 1.2 GW campus that could be built in factories and deployed faster. Then there is TerraPower, backed by Bill Gates, which is planning future projects with 690 MW reactors. This blend ensures Meta is not betting on a single technological horse but is instead securing a pipeline of supply across different timelines and maturity levels.

The strategic positioning is clear. By locking in this unprecedented volume of nuclear power, Meta is cementing its role as one of the most significant corporate purchasers of nuclear energy in American history. This moves the company from being a major energy consumer to a foundational buyer, shaping the development of the US nuclear industry. For the AI race, the winner will be the one that controls the infrastructure. Meta's deals are a direct bet on securing that control, building the energy rails to carry the next paradigm of computing.

Financial Impact and Execution Risks

The scale of Meta's nuclear bet demands a financial model to match. The company's capital commitment is staggering, with its 2025 capex expected at $72 billion. This figure is set to rise "significantly" in 2026 as the company invests aggressively in data centers. The ultimate goal-a $600 billion digital infrastructure pledge by 2028-far exceeds the company's total public market revenue over its first 15 years. Funding this requires more than internal cash flow; it necessitates external capital and innovative structures.

To de-risk its advanced reactor pipeline, Meta is using prepayments as a strategic tool. Its agreement with Oklo provides funding to advance Phase 1 of a 1.2 GW power campus in Ohio. This prepayment model serves a dual purpose: it secures future power at a known price and provides critical early capital for Oklo to procure fuel and advance project development. This approach effectively shifts some of the financial and execution risk upstream, accelerating the timeline for a technology that could be crucial for Meta's long-term power needs.

Yet the path is fraught with execution hurdles. The most significant risk is regulatory and permitting delay, particularly for advanced reactors. While the Department of Energy has launched a Reactor Pilot Program to fast-track licensing, the process for new nuclear designs remains complex and lengthy. The Oklo project's first phase is targeted to come online as early as 2030, but this timeline is ambitious and hinges on smooth regulatory approval. Furthermore, the sheer scale of Meta's promised spending introduces a capital allocation challenge. The company has already formed a joint venture with Blue Owl Capital to finance its $27 billion Hyperion data center in Louisiana, with Blue Owl owning 80%. This model of partnering with external capital may become a necessity for other projects, adding layers of complexity to its infrastructure build-out.

The bottom line is that Meta is betting its financial future on a multi-year, high-stakes infrastructure build. The prepayment strategy is smart for de-risking supply, but the company must navigate a landscape of regulatory uncertainty and massive capital requirements to turn its nuclear commitments into the power rails it needs for the AI S-curve.

Catalysts and What to Watch

The investment thesis for Meta's nuclear bet hinges on a few critical milestones. The near-term path will validate whether the company is successfully building the energy rails for the AI S-curve or facing costly delays.

The most immediate operational test is the expected launch of the Prometheus supercluster in 2026. This is not just another data center opening; it is the first major compute load powered by this new energy supply chain. The early power draw will be a key metric. It will show if the Vistra plants can deliver the promised capacity and if the advanced reactor pipeline is on track to follow. Any significant delay or technical snag here would directly challenge the timeline for the entire AI infrastructure build-out.

Parallel to this is a regulatory race. The Department of Energy's Reactor Pilot Program aims to reach criticality for at least three advanced reactor concepts by July 4, 2026. This is a hard deadline that Meta is watching closely. The program's success-or failure-will signal whether the advanced nuclear industry can accelerate from concept to construction. For Meta, which is funding projects like Oklo's, this timeline is a critical de-risking event. It will determine if the prepayments made today can translate into real power within the decade.

Finally, investors must track Meta's progress in securing additional power agreements and the financial impact. The company's agreements could end up totaling more than 6 gigawatts, but the deal with Vistra is just one piece. The financial model relies on locking in more capacity at stable prices. The balance sheet impact of this massive capex, which is set to rise "significantly" in 2026, will be scrutinized. The partnership model used for the Hyperion data center may become a template for other projects, but it also introduces new financial and operational layers. The key will be whether Meta can secure this power without straining its capital allocation or sacrificing the long-term strategic advantage of being a foundational buyer.