The energy dilemma behind the scramble for nuclear power by tech giants: Under AI pressure, U.S. power facilities are nearing their limits

When Meta announced a nuclear power procurement agreement on June 3, the market saw a harsh reality: the U.S. power infrastructure is being pushed to its limits by AI demand, and the existing 94 nuclear reactors cannot meet this energy famine.

Under the AI wave, the demand for data centers is rapidly disrupting the grid system, with tech companies scrambling to secure power to run AI systems. This exposes a fundamental mismatch with the energy industry, which typically requires years or even decades to plan and build new power plants.

"In the coming years, there will simply not be enough power to satisfy the tech industry," said Adam Stein, director of the nuclear innovation project at the Breakthrough Institute:

“Building new nuclear power plants is technically feasible, but it requires sufficient pre-planning and a complete supply chain. Unfortunately, we don't have that right now.”

Energy Panic Under the AI Wave

The last large-scale nuclear power plant in the U.S. broke ground and became operational over a decade ago. Even if a few closed plants were restarted and existing facilities upgraded, it would still not generate the amount of electricity needed for AI.

U.S. electricity demand is expected to soar by 15% in certain areas by the end of this decade, with data centers being a key driver. Texas alone has indicated that it needs the equivalent capacity of 30 traditional nuclear reactors to meet the growing demand for AI, which would easily exceed the grid's capacity.

Syed Bahauddin Alam, an assistant professor of nuclear, plasma, and radiological engineering at the University of Illinois Urbana-Champaign, warned:

"The accelerated development of AI is pushing our energy infrastructure to its limits."

Alam pointed out that what is happening now is the first wave of AI. Algorithms developed in the early 2020s are now being deployed, aimed at testing and refining the thinking and problem-solving capabilities of models. This process could last until 2030 to 2035 and will require a significant amount of energy. More nuclear power may emerge, but it will have to wait until "the tail end of the first wave."

Nuclear Power Race Among Tech Giants

Tech companies have begun frantically seeking partners to secure nuclear power supplies.

On Wednesday, media reports indicated that Meta signed a 20-year nuclear power procurement agreement, under which, starting in mid-2027, Meta will take on the entire output of approximately 1.1 gigawatts from the Clinton nuclear power plant in Illinois, enough to supply electricity to about 1 million households.

More ambitiously, Constellation is considering plans to build a second reactor at the Clinton nuclear power plant, which has received federal approval for the construction of a second unit. Constellation's CFO Daniel Eggers stated that the company could sign more such agreements. The company has several other subsidized reactors that will expire in the coming years and is negotiating with multiple potential customers. The company also plans to upgrade several existing plants, which could increase capacity equivalent to an entire new reactor:

"We have a lot of available capacity, and if customers want to procure that output, we can do it."

In March of this year, tech giants including Amazon, Google, and Meta signed a commitment led by the World Nuclear Association, calling for a doubling of global nuclear energy by 2050. This unprecedented collective action exposes the concerns of tech giants about the energy bottlenecks behind the AI computing power race.

The First Wave of Demand: The "Impossible Task" of the Nuclear Power Industry

But at least in the next few years, these efforts may not be enough. Analyst Chris Gadomski warned at an event in April that the slow project deployment and the current lack of construction indicate that nuclear power may miss the first wave of electricity demand from hyperscale data centers.

The construction costs of nuclear reactors are extremely high, and even next-generation cheaper models require government support. According to media reports on the 24th of last month, Trump signed an executive order planning to initiate the construction of 10 large nuclear power plants by 2030 and quadruple U.S. nuclear power capacity by 2050. This week, details of the Trump administration's budget request for fiscal year 2026 revealed that it plans to use $400 billion from a green bank to provide billions of dollars in loans for projects including nuclear reactors.

Nevertheless, even if a large new nuclear power plant were to start construction immediately, it could take years to deliver power to data centers.

"Nothing can help them cope with the first wave of AI demand," summarized Stein from Breakthrough Institute. This energy famine is racing against the timeline for infrastructure construction, and the outcome is already apparent