The recent artificial intelligence-related Nobel Prize in chemistry is just more evidence that the AI revolution is well underway, but there will be unobvious barriers to overcome. One will be the almost unimaginable amount of energy required to support some of its most important applications. For example, 50 pending requests for the creation of 90 new data centers in Ohio would triple the peak load on Ohio American Electric Power’s grid compared with 2023, enough to power over 20 million households.
For context, consider that each typical AI data center will consume annually the equivalent electricity of 750,000 electric vehicles — about six months of Tesla sales. The requested 90 data centers in Ohio are the power equivalent of 45 years of current Tesla sales. The numbers that support this calculation are readily available and are based on an average of 10,000 miles of driving per year.
Put another way, the power demands for EVs are child’s play compared to the electricity gluttony of AI computing, which is the power infrastructure challenge of our time. This critical matter was not getting the attention it deserved until recent weeks, when corporate titans Google and Microsoft responded to the challenge by embracing nuclear power as the future. Google is now providing funding to its partner, Kairos Power, to move ahead aggressively with its nuclear power plans, and Microsoft has announced a 20-year commitment to purchase power from a recommissioned Three Mile Island reactor in Pennsylvania. Other companies are now likely to follow.
It’s no wonder. Colocating or “near-locating” data centers with nuclear power plants eliminates many of the infrastructure costs and challenges associated with obtaining power from distant facilities.
Transitioning to small-scale nuclear reactors that can be located at or near data centers is feasible. One reactor the size of those that power a modern aircraft carrier could run two or three data centers. Considerable research into this technology has already been done by leading scientists such as Jacopo Buongiorno at the Massachusetts Institute of Technology, our alma mater. And if necessary, the Navy can be conscripted into aiding this initiative.
There is much engineering to be done to standardize and mass-produce microreactors with state-of-the-art passive safety systems to virtually eliminate serious mishaps and allow for simplified installation. The result must be closer to “plug and play” than on-site, one-off construction that has been the standard for nuclear plants in the past. Facilitating the new small-scale nuclear facilities is a goal that has already been delayed far too long.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
For those who believe renewables are the answer to the continuous power needs of a data center, think again. We calculate that the Ohio situation would require 30,000 wind turbines occupying 40,000 acres at a cost of some $40 billion. For solar, the acreage would be at least five times greater than wind at triple the cost. Throw in battery backup for about $50 billion-70 billion per day of operation during nights, clouds, or calms, plus transmission facilities to the suitable acreage, and it is clear that renewables are ill-suited to the burgeoning power demands.
Bottom line: It is in the interest of utilities, computing providers, and government agencies to form an alliance to commercialize microscale nuclear plants rapidly, assure designs with proper safety and maintainability, and develop the manufacturing, installation, and operations technologies to drive down costs. There is no time to waste if the promise of AI is not to be shackled by insufficient power grid capacity.
Andrew I. Fillat spent his career in technology venture capital and information technology companies. He is also the co-inventor of relational databases. Henry I. Miller, a physician and molecular biologist, is the Glenn Swogger distinguished fellow at the American Council on Science and Health. They were undergraduates together at MIT.