The race to build ever larger artificial intelligence models is colliding with a hard physical limit: electricity. As data centers multiply and power-hungry chips proliferate, Big Tech is turning to nuclear reactors, both old and new, to keep the lights on. In the process, companies that once branded themselves as clean-energy champions are embracing one of the most controversial technologies in the modern energy mix, gambling that nuclear power can be scaled faster than the political and safety risks that come with it.
Behind the glossy renderings and climate rhetoric, this is a high-stakes experiment in rewiring the grid around corporate AI ambitions. The sums involved, the public guarantees required, and the communities affected suggest that the nuclear turn is less a tidy climate solution than a sprawling industrial bet whose costs and benefits will be unevenly shared.
The AI energy crunch that pushed Big Tech to nuclear
The starting point is simple: AI is devouring electricity at a pace that traditional planning never anticipated. Total AI hyperscaler capex is projected to hit $530 billion in 2026, up from roughly $400 billion last year, a surge that is driving a parallel boom in power infrastructure. Global data center investment is rising in tandem, and I see that capital wave pulling utilities, regulators, and equipment makers into the orbit of a handful of cloud giants that now dictate where and how new generation gets built.
That buildout has real-world consequences. According to the IMF, current energy consumption by data centers globally already equals that of the country of France, which has a population of more than 68 million people. As AI workloads grow, grid planners warn of looming shortfalls, and states like New Mexico are openly asking who will pay for the new plants and transmission lines that hyperscale computing requires. In that context, the pivot to nuclear looks less like a bold climate play and more like a scramble for firm, around-the-clock power that can be contracted for decades at a time.
From SMRs to megaprojects: how Amazon, Google and Meta are moving
Amazon and Google have become the most visible champions of next-generation nuclear, particularly small modular reactors, or SMRs. Both companies are steering clear of building legacy gigawatt-scale reactors and instead are betting that factory-built modules can be deployed faster and closer to their data centers. Google has touted one of its agreements as the world’s first corporate deal to directly support a commercial SMR project, a signal that it wants to be seen as a first mover in nuclear-backed cloud computing rather than a reluctant buyer of whatever power the grid can spare.
Amazon is going further by tying its brand directly to specific reactor designs. The company has released new images of a proposed SMR plant, highlighting a partnership structure that pairs its cloud business with nuclear developers. It has also signed agreements with X-Energy and Dominion Energy in Virginia to add small modular reactors that would feed regional data centers and, in some cases, directly power its operations. Meta is taking a different route, lining up a massive supply of nuclear power through agreements with TerraPower and Oklo and Vistra its Prometheu data center complex in New Albany, Ohio, rather than sponsoring a single flagship SMR.
Three Mile Island’s revival shows the public is on the hook
While tech executives talk up modular innovation, the most concrete nuclear play tied to AI so far is a decidedly old-school plant: Three Mile Island. Constellation Energy Corp expects to restart the Three Mile Island plant earlier than originally planned, targeting 2027 to bring the reactor back online. The facility, on an island in the Susquehanna River near Middletown, had been out of operation for five years when Constellation Energy announced it would spend $1.6 billion to restart Unit 1, explicitly linking the project to soaring demand driven by data centers.
That restart is not being financed by tech money alone. President Donald Trump has backed the Three Mile Island restart with a $1 billion loan from the Energy Department, a federal guarantee that shifts a large share of the risk onto taxpayers. Constellation has rebranded the project as the Crane Clean Energy Center and, as it marked the one-year anniversary of the restart announcement on a Saturday in Sep, it framed the plant as essential to the digital economy and U.S. competitiveness. The message is clear: nuclear power is being revived as a strategic asset for cloud computing, with public credit and political capital underwriting the experiment.
Subsidies, SMR hype and the unresolved safety debate
Behind the scenes, the nuclear sector’s finances remain fragile. Analysts note that the nuclear power sector cannot survive without subsidies and public guarantees, and that is precisely what the tech industry is counting on as it signs long-term contracts for future reactors. A detailed assessment in Jan argues that nuclear power sector is leaning heavily on government-backed loans, tax credits, and price supports, raising the question of whether AI-driven demand is being used to justify a broader bailout of an industry that has struggled to compete with cheaper renewables and gas.
Proponents counter that nuclear energy has almost zero carbon dioxide emissions at the point of generation, a trait that makes it attractive for decarbonizing grids that are already saturated with intermittent wind and solar. A recent analysis of Nuclear options for AI data centers notes that reactors can provide stable baseload power but also stresses that they create nuclear waste that must be managed carefully and face high upfront costs and long construction timelines. Critics go further, arguing that Big Tech and Big Tech and nuclear boosters are pushing small modular reactors as a climate solution, but that this is incredibly misleading because SMRs still carry accident and proliferation risks while diverting capital from faster, cheaper clean energy. In that light, the AI-nuclear nexus looks less like a straightforward decarbonization strategy and more like a contested bet on a technology with unresolved safety and waste legacies.
Who pays, who benefits and who bears the risk
As more AI-linked nuclear projects surface, the central political question is who ultimately pays. In New Mexico, policymakers wrestling with the AI boom’s power demands have warned that Nuclear investment remains risky and heavily dependent on regulatory certainty, with Early performance of next-generation reactors still unproven at scale. A separate assessment of the AI Energy Challenge notes that Big Tech Is Betting on long-lived nuclear assets that will outlast any single generation of AI hardware, raising the possibility that communities and ratepayers will be left holding the bag if demand projections fall short or technologies shift.
There is also a geographic and social justice dimension that is easy to miss in corporate sustainability reports. According to campaigners tracking the AI buildout, many of the proposed nuclear and fossil plants that would serve data centers are sited near communities that already live with industrial pollution, including so-called “downwinders” and Indigenous groups who have long borne the brunt of uranium mining and weapons testing. As one critical report on Silicon Valley’s AI expansion puts it, According to activists, the AI boom is becoming an environmental time bomb for these communities, even as executives tout clean energy partnerships on earnings calls. When I look at the emerging map of nuclear-backed data centers, from the Crane Clean Energy Center in Sep to the new complexes rising near New Albany, it is clear that the AI revolution’s power supply is being built on a foundation of public risk, private gain, and unresolved questions about how safe, how clean, and how democratic this nuclear experiment will really be.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
