Artificial intelligence wants to develop nuclear weapons. Will it work? : NPR

MIDDLETOWN, Pa. — Standing in front of one of Three Mile Island’s iconic cooling towers, Brian Hanson remembers the day the nuclear power plant was shut down in 2019.

“It was really depressing,” said Hanson, chief power generation officer at utility company Constellation.

Three Mile Island is best known for the partial meltdown of its Unit 2 reactor in 1979. But the Unit 1 reactor continued to operate for decades until Hanson finally had to shut it down.

“I remember seeing no steam coming out of the tower and driving away thinking it would never happen again,” he says.

And earlier this year, Three Mile Island was given a second lease of life. In September, Microsoft signed a power purchase agreement with Constellation, guaranteeing that it would buy power from the plant at a fixed price. This allowed power companies to invest the $1.6 billion expected to restart the reactors.

Across the technology sector, companies like Microsoft, Google, Amazon, and Meta are getting into nuclear technology.

Their interest is driven by their efforts to combat climate change, said Emma Strubel, a computer scientist at Carnegie Mellon University in Pittsburgh. By 2022, big tech companies will have “actually reduced their carbon footprint year over year,” Strubel said.

However, since the advent of AI, power consumption has increased rapidly.

Training and using AI requires significantly more computing power than traditional computing, “which translates into energy usage,” Strubel said. Strubel and other experts predict that emissions will skyrocket as AI becomes more widespread.

Nuclear power provides a solution. Power plants like Three Mile Island can provide hundreds of megawatts of electricity without emitting greenhouse gases. New nuclear power plants have the potential to do even more, including powering data centers using the latest technology.

But the Silicon Valley ethos is to move fast and break things. Nuclear power, on the other hand, has a reputation for being very slow because nothing breaks down.

This has led to skepticism from some longtime observers of the nuclear industry.

“It’s interesting that big technology companies are interested in these technologies,” said Sharon Scassoni, a research professor at George Washington University who has studied the nuclear industry for years. But she wonders if companies realize how much time and money it takes to harness the power of atoms.

“I’m really confused,” she says.

Climate change commitments

Big tech companies have long promised sustainable growth, and in recent years have pledged to reduce or even eliminate greenhouse gas emissions.

Initially, they tried to do it using solar, wind, and hydroelectric power. “If you go back 10 years, they were all focused on 100% renewable energy,” said Ted Nord, executive director of the Breakthrough Institute, an environmental think tank based in Berkeley, California. House says. Technology department.

Big tech companies have bought electricity from renewable sources, but purchase contracts often don’t cover some demand, leading to criticism, Nordhaus said. For example, purchasing electricity from a solar power plant does little to reduce emissions during nighttime operations.

For years, some companies have struggled to match their demand with the supply of clean energy on a 24/7 basis. Nordhaus says the changes, along with the predicted huge power requirements of AI, have made nuclear power one of the few solutions.

“I think nuclear power is probably the most cost-effective stopgap technology we have today,” Strubel agrees. Because wind and solar are too intermittent, “the size of batteries that would need to be built next to the data center to support these workloads would be enormous and incredibly expensive.”

Given pressing energy needs, paying to restart factories like Three Mile Island seems like a bargain for a company like Microsoft.

“We’re talking about data centers that are extremely power-hungry 24/7, no matter what the sun is shining or the wind is blowing,” said the 36-year-old, who trained as a nuclear engineer. Hanson says. “It’s a perfect fit for nuclear energy.”

Hanson said Constellation plans to have the plant, renamed Crane Clean Energy Center, operational by 2028.

nuclear recession

If technology requires the nuclear industry, the nuclear industry may need more technology.

“There haven’t been any nuclear power projects going on for decades,” says Edwin Lyman, who tracks the industry for the Union of Concerned Scientists.

Only two power-generating nuclear reactors have actually been built in the United States in recent years, but they were billions of dollars over budget and far behind schedule. “The final cost was more than double the original estimate, and it took twice as long to get the reactor up and running than originally predicted,” Lyman said.

Meanwhile, existing nuclear power plants are struggling to keep up with cheaper natural gas and renewable energy supplies. In fact, that’s why Three Mile Island had to close in the first place. “Electricity prices started to fall and nuclear power plants, which require intensive operation, became uneconomical,” Hanson said. “We owe it to our shareholders to make those difficult decisions.”

Microsoft’s efforts have led to a turnaround. Hanson is currently busy hiring workers, inspecting old equipment and working with regulators to get the plant back up and running. “Microsoft has purchased the power for this unit for the next 20 years, giving us financial certainty to invest our money,” he says.

The smaller the better

But the solution to restarting old factories is limited because there aren’t that many old factories in the U.S. waiting for an influx of new capital. To meet future demand, technology companies are looking to build new nuclear power plants, but those reactors will look very different from older reactors.

In a commercial office district on the outskirts of Washington, D.C., a company called X-energy is promising to develop a new kind of nuclear power.

“X-energy has developed a design that is very different from what most people think of for very large conventional nuclear power plants,” said Clay Sell, CEO of the company.

Instead of using fuel rods filled with nuclear fuel, X-energy plants use small balls of uranium. “If you put round pebbles the size of a cue ball and fill the core with them, the pebbles flow through the core like gumballs in a gumball machine,” he says.

Reactors will become smaller and more modular, with multiple units powering data centers, Sell said. And importantly, he says, the system has one big advantage.

“It’s impossible for a nuclear power plant to have a meltdown in any scenario you can imagine,” he says.

In the past, companies like X-energy would have a hard time finding people to buy their designs, but that’s no longer the case. Earlier this year, Amazon invested more than $250 million in the company as part of a $500 million venture investment round. Other small reactor companies are also gaining interest. Open AI CEO Sam Altman has invested heavily in a company called Oklo, and Google has committed to buying power from a company called Kairos. Earlier this month, Meta announced it would also explore partnerships with nuclear power companies.

“Nuclear power is not only one of the few solutions, it’s clearly the best…and I think technology companies have come to that conclusion over the last couple of years,” Sell said.

X-energy says its first reactor could be operational as early as 2030, but its design still needs approval from the Nuclear Regulatory Commission.

That process will take time, and AI needs power now. Some longtime nuclear skeptics question whether Silicon Valley executives really understand what they are trying to do.

“They are champions of innovation who believe that any technology can be disrupted,” Lyman said. But “nuclear power is a tough nut to crack.”

Squassoni worries that nuclear power is simply not suitable for combating climate change.

“You want the maximum return in the short term, and that’s not the case with nuclear,” she warns. “If you’re putting money into it, you’re pushing us into deep climate change.”

Until these plants are up and running, many AI data centers will have to run using natural gas, she says.

Mr. Squassoni believes the slow nuclear industry will never be able to catch up to the fast pace of Silicon Valley, and if it can’t, other solutions will be needed. “I don’t know,” she muses. “Maybe the answer is to use less AI.”