Our AI Dreams Require Nuclear Power
Palantir and The Nuclear Company team up to usher in a new nuclear age for our AI dreams
There’s a new marriage in the world of infrastructure and software, and it reads like something from a techno-nationalist fever dream: Palantir and The Nuclear Company are teaming up to build a real-time, AI-driven Nuclear Operating System (NOS) designed to scale nuclear power construction across the United States. This partnership arrives draped in strategic urgency, national security rhetoric, and the narrative backdrop of China building 10 GW of new reactors annually while the U.S. has managed just 2 GW in the last three decades.
Palantir, known for deploying Foundry in military logistics, pharma supply chains, and covert surveillance, is now training its orchestration software on one of the most complex, failure-prone undertakings in modern industrial history: building nuclear reactors on time and on budget.
The problem is real. The electricity needed to train and run large models is running up against grid constraints. Transmission is lagging. Permitting timelines are glacial. New peaker plants face turbine supply bottlenecks through 2028. AI accelerationists preach exponential takeoff curves, but the reality on the ground looks like a 1940s war mobilization effort stuck in molasses. Nuclear should be the silver bullet: dense, carbon-free baseload. But it has repeatedly proven unscalable in the West, primarily due to construction delays, cost overruns, and regulatory drag.
So NOS enters with the promise of turning this slow, baroque art form into something closer to just-in-time precision engineering. Real-time data from sensors. Digital twins of build sites. AI agents that track permitting documents and re-route labor tasks in response to weather or material shortfalls. Instant feedback loops replacing bureaucratic opacity. If this works, it’s transformative.
But it won’t work. At least, not fully. Not yet.
To understand why, we need to ask: what actually kills nuclear projects?
It’s not lack of software. It’s:
Design churn after construction begins, often driven by shifting regulatory interpretations.
Catastrophic vendor delays for one-of-a-kind components like containment vessels and control rods.
Labor productivity collapse from a workforce unfamiliar with nuclear-grade standards.
Multi-layered permitting structures across federal and state agencies, each with its own veto points.
Financing conditions that treat nuclear as a political risk, raising cost of capital to unsustainable levels.
NOS can clean up a lot of the mid-level chaos. But it can’t conjure a new domestic heavy-forge industry. It can’t shorten NRC statutory timelines. It can’t train and certify 100,000 nuclear-qualified tradespeople. And it can’t make bankers believe a 15-year build with no guaranteed offtake is worth funding without federal backstops.
So what’s the point?
Despite its limits, NOS matters. Because if America is going to build 400 GW of nuclear by 2050, as called for in the latest Trump executive order, we will need industrial orchestration software just to keep the effort from descending into chaos. NOS won’t eliminate risk, but it can make risk legible. It offers three nontrivial benefits:
Standardization Flywheel: If The Nuclear Company can really deliver on a “design-once, build-many” model (think SMRs or Gen III+ templates) then a unified data layer becomes a strategic asset. Each site builds on the last. Quality assurance artifacts, vendor qualifications, and regulatory feedback form an accretive base. That’s how China builds. It’s how chip fabs scale. America’s historical nuclear approach, by contrast, has been artisanal and bespoke.
Supply Chain Telemetry: Foundry has already shown it can surface bottlenecks in defense supply chains. Apply that to nuclear: if one supplier of pressure valves is causing monthslong delays across four sites, you now have a justification for targeted DoE or DoD industrial base support. Without telemetry, every delay looks like a local failure.
Political Cover: This is underrated. A software system that gives regulators and funders real-time insight into schedule risk, material availability, and quality assurance status makes the entire ecosystem more governable. It gives elected officials and program managers an excuse to keep writing checks. If you want a Manhattan Project for energy, you need dashboards.
So NOS is a partial answer to a real problem. And it aligns with broader moves: hyperscalers are beginning to look at on-site SMRs as baseload anchors for compute clusters. The Trump wing of the GOP is embracing nuclear nationalism as a kind of geopolitical reindustrialization.
But we should be sober about where NOS fits in the hierarchy of constraints. It is not the keystone. It is an amplifier. The real gating functions remain:
Heavy-forge capacity and component availability
Regulatory streamlining, especially NRC Part 53 reform
Labor mobilization and training
Federal credit support to lower cost of capital
If those pieces move, NOS will make the system faster, more predictable, and more auditable. If they don’t, NOS will be the smartest conductor of an orchestra that never arrives.
The deeper truth is this: the AI era needs nuclear power, and nuclear power needs an industrial base revival that goes far beyond software. But it also needs software to make that revival tractable. NOS isn’t the breakthrough we’ve been waiting for. But it might be a necessary precondition for the breakthroughs to come.
Coda
If you enjoy this newsletter, consider sharing it with a colleague.
Most posts are public. Some are paywalled.
I’m always happy to receive comments, questions, and pushback. If you want to connect with me directly, you can:
was just talking about this with a european diplomacy person, nuclear and ai defense are hand and hand against Chinese perversion of global influence.