Idaho National Lab and NVIDIA Target 50% Cut in Nuclear Reactor Build Times
The Prometheus initiative carries $293 million in DOE backing and a claim that AI can compress construction schedules that have bedevilled the industry for decades.
The URA uranium exchange-traded fund dropped 3.79% to $43.62 as of August 19 (2026-08-19), trimming gains in a sector that has drawn heavy capital on the back of AI-driven power demand. The pullback came as attention turned to one of the more specific AI propositions in nuclear to date: Idaho National Laboratory has partnered with NVIDIA to launch Prometheus, a program using artificial intelligence to target a 50% cut in reactor build times and a similar reduction in operating costs, according to carboncredits.com. The U.S. Department of Energy has committed $293 million through a competitive program to support it.2
The ambition sits in tension with history. Nuclear construction in the West has been characterised by overruns rather than compression. American and European projects over the past two decades delivered late and over budget by multiples, not margins. The Prometheus program's own milestones and verification criteria were not detailed in available reporting, which makes the 50% figure a target rather than a demonstrated result.2
The nuclear sector's AI moment extends beyond U.S. federal labs. During the week of July 20 (2026-07-20), researchers at the Chinese Academy of Sciences presented a plan at the World Artificial Intelligence Conference in Shanghai for integrating AI across the full nuclear energy lifecycle, from design through operations, oilprice.com reported. The Chinese proposal placed particular emphasis on safety — on how AI systems can be embedded in environments where failures are not recoverable and where regulatory scrutiny is intense.4
That question of safety validation is sharper in nuclear than in almost any other industry. Regulators license nuclear facilities against deterministic engineering standards; an AI system's probabilistic outputs and the opacity of its decision logic have not been accommodated by existing frameworks in most jurisdictions. The CAS proposal did not specify how Chinese regulators would be brought along.4
On the maintenance side, the economic case for AI is more direct. A single day of unplanned downtime costs a reactor operator roughly €1 million in lost electricity production, according to power-technology.com, making predictive maintenance a defensible application well before AI approaches anything as sensitive as reactor control systems.3
The IEA has described the current nuclear moment as carrying "fresh momentum" with "the potential to open a new era," attributing renewed interest partly to energy security concerns that intensified after Russia's invasion of Ukraine. That assessment is cautious for a reason. The agency has noted renewed interest before; whether capital, regulatory reform, and construction capacity align this time is separate from whether the technology works.3
In public equity markets, the gap between sentiment and balance sheets is visible. Babcock & Wilcox, which makes industrial power equipment and has pivoted toward data-centre baseload, saw shares rise 129.34% year to date to $14.54 as of May 21 (2026-05-21). The company secured a $2.4 billion design-build contract with Base Electron for 1.2 gigawatts of natural gas-fired capacity, driving backlog up 470% to $2.8 billion. Management guided 2026 core adjusted EBITDA at $70 million to $85 million, representing roughly 80% year-on-year growth, with no data-centre revenue included in that figure.1
Set against those numbers, Babcock carries stockholders' equity of -$131.5 million and faces refinancing on a 6.50% note due in 2026. Base Electron is evaluating an additional 1.2 GW option, and the global project pipeline is reported above $12 billion. But the balance sheet condition means a delayed draw on that pipeline or a refinancing stumble lands differently at Babcock than at a company with a stronger equity base.1
The Prometheus project's core claim — that AI can halve reactor development timelines — will face its first real test not in Idaho but in Washington. The Nuclear Regulatory Commission has not established a pathway for validating AI tools in safety-critical nuclear applications. Until it does, the $293 million in DOE funding accelerates the engineering. Whether it accelerates the licensing is a different matter entirely.2