Major tech companies should invest in building nuclear power plants to meet AI energy demands

What's this about?

People disagree about whether big tech firms should help build nuclear power plants for growing AI use.

The key question asks if nuclear power can meet future needs safely, cheaply, and on time.

What supporters say

• AI may need much more power, so tech firms may want new power sources they can count on.
• Nuclear plants can make steady power day and night, with low carbon pollution.
• Tech deals could help new reactor designs become real businesses, though success remains unclear.

What critics say

• Better computer tools and flexible AI work may lower the total power needed.
• New nuclear plants may take too long to help with power needs soon.
• Building plants can cost far more than planned, making them poor buys.

How to read this

The number of points on each side does not show who is right; strong proof matters more than many points.

The bottom line

The facts give tech firms a good reason to support nuclear power.

But the facts do not yet prove they should own or fund new plants without strong safety rules and a wider power plan.

Major technology companies are weighing whether to do more than buy electricity for their expanding AI data centers: they could help build or finance nuclear plants. The evidence gives them a credible reason to support nuclear power, but it does not clearly justify owning or funding new reactors without strong safeguards and a broader energy strategy.

The case for

Nuclear power offers something AI data centers increasingly need: steady electricity with relatively low carbon emissions. Unlike wind and solar, reactors can operate around the clock, making them useful for facilities that run at high capacity. The International Energy Agency identifies nuclear as one possible source of firm, low-emissions power for growing data-center demand, while the IPCC classifies nuclear as low carbon over its full lifecycle (see Figure 2). 1

The expected growth of AI also gives technology companies a strategic reason to support new firm generation. Lawrence Berkeley National Laboratory estimates that U.S. data centers already use a significant share of the country’s electricity and could consume considerably more by 2028. The laboratory also stresses that the forecast depends heavily on uncertain assumptions about AI growth and efficiency. That uncertainty could make long-term contracts or project financing valuable, especially if companies want more control over future power supplies. 2

Corporate commitments might also help advanced nuclear designs move from demonstrations toward commercial deployment. The Energy Department has identified data centers as potential early customers for advanced reactors. Google has announced an agreement involving up to 500 megawatts from multiple advanced reactors, while Microsoft has made an arrangement linked to restarting an existing plant. These deals show that technology companies can create demand, although they do not prove that new reactors will be built or operate successfully. 3

The case against

The strongest objection is timing. AI electricity demand may rise faster than new nuclear plants can be planned, licensed, financed, connected to the grid and built. The IEA reports delays, cost overruns and financing problems in many new nuclear projects in advanced economies. MIT has said that shorter construction schedules and more reliable delivery are necessary for wider deployment, while the Government Accountability Office finds that commercial deployment of small modular reactors remains limited and that licensing, supply-chain, cost and demonstration problems are unresolved. 4

New reactors also expose companies to major financial risks. Nuclear costs depend heavily on construction time, interest rates and overruns. The OECD and IEA identify financing and construction duration as especially important, while the Union of Concerned Scientists points to recent U.S. projects as evidence of serious cost and schedule risks for investors and ratepayers. 5

Meanwhile, better technology could reduce the amount of dedicated nuclear capacity AI ultimately requires. Lawrence Berkeley National Laboratory says demand forecasts are highly sensitive to efficiency assumptions. Research also suggests that some AI workloads can be delayed or shifted, lowering peak demand and making it easier to combine renewables with storage and other flexible resources. 6

That makes a distinction important: buying power from an existing reactor or supporting a restart is not the same as financing a new plant. Existing facilities can provide nuclear electricity sooner. Microsoft’s arrangement concerns an existing unit, while Amazon’s proposed colocated expansion still faces regulatory and grid-allocation questions. Neither example shows that technology companies should own and build reactors themselves.

The bottom line

The evidence favors a qualified, not categorical, endorsement of corporate support for nuclear power. Nuclear can provide dependable low-carbon electricity, and rapidly growing AI demand gives major technology companies a reasonable motive to secure more firm generation. But the evidence against directly investing in newly built plants is stronger on the practical question of timing and commercial risk.

The most defensible approach is a diversified one: consider existing nuclear and restarts alongside renewables, storage, transmission, efficiency and flexible computing. Corporate commitments could be justified where there is a clear reliability need and where construction, regulatory and financial risks are openly allocated.

Overall confidence is low, despite a substantial evidence base. The research is much better at showing that nuclear can produce firm low-carbon power—and that new projects face serious risks—than at predicting whether a particular reactor will be completed on time, within budget or at the scale AI demand eventually requires.