January 31, 2025
4 I will read the smallest
AI’s energy demand threatens the nightmare of nuclear waste
Promoting AI to revive a nuclear power plant threatens the avalanche of nuclear waste

The three-mile nuclear power station on the London Delly Township, Pennsylvania, Pennsylvania, was closed on March 28, 1979 following the partial meltdown of the unit 2 reaction device (TMI-2).
Andregenny/Alamie Stock Photos
After the decline, the US nuclear industry hopes to be revived in three malls of Pennsylvania, Hanford Sight in Washington, and the biggest failure. The nuclear power that the industry claims helps to meet the rapidly increasing power demand from the data center and the growing AI economy. However, such a wrong order ignores the long -standing unresolved problem of radioactive nuclear waste that AI cannot.
In September, CONSTELLATION Energy announced a plan to restart THREE MILE ISLAND shutter creators protruding by Microsoft. Following the lawsuit, Amazon announced in November that it will invest $ 334 million to develop a small module reactor (SMR) in Hanford, the world’s first Plutonium production facility.
Google and Meta also want to regain nuclear power. In October 2024, Google eventually has a long -sophisticated Uranium for a long and sophisticated national lab for a long sophisticated national lab, and is developing a new SMR in Oakridge, Tennessee 500 Megawatts. We announced that we plan to purchase. Facebook’s parent Meta is looking for a bidding at a nuclear power plant in the data center.
About the support of science journalism
If you are enjoying this article, consider supporting journalism that has won. Subscription. By purchasing a subscription, it will help you secure the future of a story that has an impactful story that forms our world today.
These high -tech giants recognize that the next -generation microcessors used for AI calculation at the data center require a large amount of electricity to supply and cool power. For example, a single NVIDIA BLACKWELL chip can create up to 2 kilowatts than a typical house. Thousands of people are packed in the server in the data center, which requires the same power as a small city.
The so -called Hyper Scale Data Center must exceed 100 Megawatts (100 MW). This is a considerable part of the output of a major power plant. And its power must be cheap, stable, and reliable.
The prestigious report of December 2024 from the US Ministry of Energy, written by energy experts at Lawrence Berkeley National Research Institute, is particularly illuminated. The increase in energy consumption centered on US data over the next five years has stated that “it will respond to the total power demand of 74 to 132 (Gigawatt) data centers.” This represents about 7-12 % of US power consumption forecasts in 2028.
Where do all of these power come from? It is a bit strange that BIG TECH has returned to the nucleus, given the challenges of power companies supplying electricity and satisfying other growth needs, including electric vehicles. However, the power demands for the DOE reports must be built or resurrected in the next five years of at least 40 miles of nuclear reactors. It is impossible.
A few years ago, Amazon, Google, Meta, and Microsoft promised not to exacerbate the carboxyide level in the atmosphere. However, given the explosive power consumption of the data center, the goals worthy of their praise are becoming more and more difficult. As a result, they began to advertise the return to nuclear power to avoid this troublesome problem. That is a big mistake.
Nuclear power is certainly a source of energy without carbon emissions, but it is definitely not a clean energy source, but definitely do not have Playable. It is present in radioactive release, along the ulan supply chain from mining to concentration, fuel rods and pellets. For example, in South Texas, landowners are worried about the contamination of groundwater due to nearby updated uranium mining activities.
Since 1989, DOE has spent hundreds of billions of dollar taxpayers who clean the original nuclear complex, including a huge oak lidge factory that concentrates many of uraniums used in commercial nuclear power. And despite the decades of attempts, the department has not yet completely purified and dismantled the collapsed tank that oozes out the very highly radioactive waste remaining from Hanford’s plutonium processing.
The storage and containment of spent nuclear fuel is actually an important unresolved issue in the US nuclear industry. More than 90,000 tons of these waste are stored on 77 sites in 35 states, increasing by more than 2,000 tons per year.
Small modularities that Microsoft’s co -founders, Bill Gates, etc., only increase this burden. Former U.S. Nuclear Regulations (NRC) Chair Allison McFarren and Rodney C. Uing of Stanford University stated: Just more wasted, period. “
In other places, Macfarlane emphasized the procedures and actual difficulties facing new reactor technology when accepting NRC and achieving commercial success. For example, shortly after receiving NRC certification in 2023, many Nusukale SMR projects were abandoned after doubling construction costs of $ 9.3 billion or more. Aside from waste problems, commercially successful SMR designs are probably more than 10 years away.
However, a relentless AI Gold rush imposes a demand for the predicted power supply before it is not checked. On the other hand, in light of the laws of supply and demand, the power rate rises without mercy. The approaching energy crisis explains the Big Tech’s efforts to slow down fossil fuel power plants and revive the shuttered reactor.
Amazon, Google, Meta, and Microsoft executives need to take a deep breath and start re -evaluation of options. Do they really need to build and upgrade data centers at such a fierce pace? Or is this Devil Take the Hindast AI Arm Race just a bitter competition due to recent progress in semiconductor technology?
And what about the truly beautiful and renewable energy source they once accepted? How about sunlight, wind, and geothermal? Yes, the vibration of solar energy and wind energy is inadequate with the stable power requirements in the data center. However, energy storage has recently been on a long way and has a promising future. And the recent surprising success of the Chinese DeepSeek AI program indicates that software efficiency plays an important role in this initiative.
Given that dark clouds still remain in nuclear power, especially unresolved waste, these revitalized alternatives deserve new considerations.
This is an article on opinions and analysis, and the views expressed by the author and the author are not always Scientific American.