Max Headroom 
While the Defense Department is likely to benefit from OpenAI’s announcement this week that it would invest half a trillion dollars to build new artificial intelligence data centers around the country, Pentagon officials warned that the U.S. lacks the energy resources and computing power to support the new infrastructure — and solving that problem won’t be easy.
OpenAI announced the project, dubbed Stargate, on Tuesday, pledging an initial $100 billion — plus another $400 billion over the next five years — to build new AI infrastructure across the U.S. and create “hundreds of thousands of American jobs” in the process. Early funders include Softbank, OpenAI, Oracle and MGX — a technology investment firm based in the United Arab Emirates — and OpenAI will partner with Oracle, Microsoft, Arm and NVIDIA on technology development.
During a press conference Tuesday at the White House, President Donald Trump called the effort a “monumental undertaking” and said the White House would support the project, in part, through issuing emergency declarations, though he didn’t expand on details.
The Defense Department has an ambitious vision for using AI across a range of military missions, including data collection, intelligence analysis, campaigning and logistics. But running those tools and applications takes more computing power and space than DOD has access to.
Roy Campbell, deputy director of advanced computing in the Office of the Undersecretary of Defense for Research and Engineering, said Thursday that many times, bases outside of the U.S. don’t have the computing power they need to retrain new AI tools.
“In some cases, for you to be able to handle a situation a forward operating base can’t handle, you have to kick that back to [the continental United States] and use the DOD supercomputing centers that we have there,” he said during a panel at the Potomac Officers Club’s annual Research and Development Summit in McLean, Virginia.
Jeff Waksman, who’s leading an effort in the Pentagon’s Strategic Capabilities Office to develop a mobile nuclear reactor, said the strain that technologies like AI and high-power computing place on the electric grid raises questions about who should have access to data and how to mitigate the risk of blackouts.
“This is not a problem that industry or the DOD can figure out by itself. It’s about the nation’s grid as a whole,” said Waksman, who spoke on a panel with Campbell. “It’s probably the most underrated challenge of this huge $500 billion announcement.”
Waksman’s nuclear reactor program, known as Project Pele, offers one answer to that challenge: using nuclear power to source energy for AI computing.
The effort, initiated in 2019, aims to demonstrate the first-ever U.S. prototype of a portable nuclear reactor within five years. The mobile reactor, which the department estimates could deliver one to five megawatts of electrical power over a minimum three-year operating life, would support DOD’s growing energy needs by providing power to austere locations.
The Pentagon broke ground at the Project Pele test site at the Idaho National Laboratory last September and plans to begin assembling the reactor — built by BWXT Advanced Technologies — as soon as next month. The department aims to demonstrate the technology in 2026.
“It’s going to be the first generation portable nuclear reactor built anywhere in the world, outside of China,” Waksman said. “It’s very much not a paper project anymore.”
Another potential solution to the AI power problem is making processors more effective at crunching data. Steven Meier, associate director of space technology at the Naval Research Center, said his lab is exploring the use of more efficient neuromorphic processors that can be 100 times more efficient than a standard processor. Essentially, neuromorphic processors take up less space, work faster and use less energy.
“There’s huge gains to be made in terms of neuromorphic processors making AI and [machine learning] more accessible on autonomous vehicles of all shapes and sizes,” Meier said at the conference.
Battlefield Tech
C4ISRNet
[crypto-donation-box type=”tabular” show-coin=”all”]
