A man with a shiny metal model of the Concorde on his desk and old circuit boards on a shelf is clearly obsessed with technology.
Chris Sharp is Chief Technology Officer at Digital Realty, a US company at the intersection of construction and high technology. It builds data centers, anonymous warehouses full of computers that keep the online world spinning.
And with the advent of artificial intelligence (AI), which requires far more processing power than standard computing, the rocket booster is in the data center world.
His company just built a huge new data center dedicated to AI in Portland, Oregon. How different is this from a regular data center?
“A typical data center requires 32 megawatts of power flowing through the building; an AI data center requires 80 megawatts,” Sharp said.
AI systems use all of this extra power simply because they are doing much more processing than standard computing. They are crunching through far more data.
Sharp also points out that the overall web of technical support required for AI is much larger. “For example, there is five times as much cabling.”
All of these points to a problem. How can AI grow if it requires so much power to function?
Demanding ever more power from the existing power grid means competition from households and other industries, and the data center sector has no allies in the event of a power outage.
“Our industry has to find another source of power,” Sharpe declares. He thinks it’s nuclear.
More urgently, he predicts that in the not too distant future, data centers will have their own nuclear reactors.
The technology in question is the much-touted small modular reactor (SMR). These are designs for advanced nuclear reactors that produce about one-third as much power as traditional large nuclear power plants.
Although there are currently no SMRs in commercial operation worldwide, China is building the world’s first SMR, and similar technology is already used in nuclear submarines.
Meanwhile, universities such as Britain’s Imperial College London have been operating small nuclear reactors for education and training purposes for years. Imperial’s nuclear reactor, located outside London, operated from 1965 to 2010.
Most companies currently developing commercial SMRs are focused on helping keep the lights on in towns and cities. However, many professional companies have determined that data centers are the best candidates for SMR designs.
Dr Michael Black runs the Center for Nuclear Engineering at Imperial College London. “Data centers are power hungry, but the introduction of AI is moving power requirements to a new level,” he says.
“There are about 50 SMR designs out there. The challenge is to build them in repeatable units, factory style, and standardize the production line.
“Apart from having to pass regulations, there is no reason why small fast reactors cannot power data centers.”
In the United States, one SMR design from a company called NuScale has already been given the green light by the Nuclear Energy Administration. Meanwhile, in the UK, the Nuclear Regulation Authority continues to study SMR designs from Rolls-Royce and US tech company Holtec International.
US energy company Westinghouse also wants to build four SMRs in the Tees Valley in north-east England, close to the existing Hartlepool nuclear power station.
Read additional articles about artificial intelligence
Dr. Brooke believes that nuclear submarines provide a compelling safety case for SMR technology. “If you build a nuclear reactor and put it on a submarine, people will sleep within two yards of the reactor.”
And he predicts that the technology sector’s involvement in SMR development will be decisive. “These guys have a lot of money!”
By contrast, Dr. Doug Parr, chief scientist at Greenpeace UK, says the high cost of SMRs will be too big a barrier.
There is unrealistic hype behind SMR cost estimates. “This hype will dissipate as delays and difficulties surface,” he says.
Dr. Parr argues that data center operators will become indifferent if costs are revised to the point where SMR becomes uncompetitive compared to renewable energy sources.
Greenpeace also opposes nuclear power generation on safety grounds, pointing out the risk of accidents and the need to deal with radioactive waste.
Spencer Lamb, chief commercial officer at British data center developer Kao Data, doesn’t think a UK nuclear facility is imminent. “We’ve heard about SMR, but it will take a long time to roll out a nuclear-configured data center in the UK,” he says. “And AI is happening now.”
Back in California, a company called Oklo announced that its SMR design was nearly complete.
“AI is the catalyst, the main driver,” said Brian Gitto, head of business development at the company.
Oklo plans to manufacture SMRs, saying they can be built quickly and safely.
“People are questioning the viability of nuclear power because of the waste and the risk of accidents,” Gitto says. “We recycle the fuel over and over again in the reactor and dispose of the waste. New reactors don’t melt down, they’re self-cooling, they’re self-regulating.”
Gitto declined to say whether any data center companies have already signed up, but said “all major operators are interested.” He added: “Power is their biggest problem and they want to have this technology in place within four years, by 2028.”
Gitto says his 25 years in the energy business have convinced him that nuclear power is the only solution to the pressures facing the new AI data center world.
”[Without us] You just don’t have permission to power on all the machines you need. We have signed a letter of intent for specific data center locations for his 2028 period. ”
So what would a dedicated nuclear power plant for a data center look like? Oklo envisions dropping a large stainless steel tank housing an SMR into a hole 50 feet (15 meters) deep in the ground. There is. Data centers will then be built on and around it.
Perhaps the ultimate alliance between SMR, AI, and data centers is evident on Oklo’s board of directors.
Sam Altman, a high-profile leader at leading AI company OpenAI, has been chairman since 2015. As Gitto says of Altman, “He identified this problem 10 years ago.”
