- The future of artificial intelligence depends on nuclear energy.
- The future of nuclear energy depends on whether America’s regulator can still say “no.”
- That question may determine who really governs the AI age.
There is both good news and bad news coming out of America’s nuclear regulator.
The good news is that the U.S. Nuclear Regulatory Commission appears finally to be evolving its thinking on artificial intelligence.
For more than two years, I have argued publicly that the commission was making a fundamental category error by regulating AI largely as though it were simply another software component operating inside a nuclear facility. The NRC maintained that its existing technology-neutral regulatory framework was generally sufficient, and much of the analyst community dismissed the argument that AI represented an entirely new regulatory challenge.
Today, the commission’s own actions tell a different story.
The NRC is now investing in AI-specific research into cybersecurity, verification, validation and regulatory gaps. It is building internal expertise, convening technical workshops and examining how adaptive AI systems should be evaluated before they are entrusted with safety-critical nuclear applications.
That is encouraging.
It is not a complete reversal of the NRC’s earlier position, but it is unquestionably a significant evolution. More importantly, it is an implicit acknowledgement that artificial intelligence behaves differently from traditional deterministic software. It learns differently. It changes differently. It fails differently.
Ipso facto, safety-critical AI deserves safety-critical regulation.
Unfortunately, this welcome evolution is occurring at precisely the wrong political moment.
The weakest player
While the NRC is becoming more technically sophisticated about artificial intelligence, it is simultaneously becoming less politically sovereign.
The Trump administration has proposed reducing the NRC’s 2027 budget by approximately 8% while eliminating nearly 200 positions from an agency that already operates on less than $1 billion annually — a remarkably modest budget by Washington standards. Those reductions come as the commission is expected to oversee 94 commercial nuclear reactors while licensing an entirely new generation of advanced nuclear technologies.
The administration has been unusually candid about its reasoning. It argues that the NRC’s “outdated and overly risk-averse” approach is preventing America from deploying the reactors needed to support “energy-intensive industries,” including artificial intelligence.
Think carefully about what that means.
For 70 years, the NRC existed to ask difficult questions before new technologies entered America’s nuclear infrastructure.
Today, caution itself is increasingly portrayed as the obstacle.
That is not merely regulatory reform. It is a profound change in regulatory philosophy. Curiously, the budget reductions achieve remarkably little from a fiscal perspective. More than 80% percent of the NRC’s operating budget is already recovered directly through license fees paid by the utilities that own and operate America’s nuclear power stations. Reducing the commission’s appropriation therefore saves taxpayers comparatively little.
If cost reduction is not the principal objective, what is?
The administration’s own language suggests another answer.
It believes the NRC has become too cautious, too slow and too independent.
The practical consequence of reducing its resources is unavoidable: a regulator with fewer engineers, fewer inspectors and less technical capacity is simply a weaker regulator.
My own conclusion is that this is no accident. The administration is not merely attempting to accelerate nuclear deployment. It is reducing the independent capacity of the NRC to challenge that deployment. Whether one agrees with that objective or not, its effect is unmistakable: a regulator with diminished resources possesses a diminished ability to regulate.
A regulator cannot regulate effectively if it no longer possesses the institutional capacity to say “no.”
Three centers of power
This story is not really about reactors. It is about sovereignty.
Today, three institutions are competing to shape America’s nuclear future.
The White House exercises power through executive orders and industrial policy.
The hyperscalers exercise influence through lobbying, campaign finance, trillion-dollar market capitalizations and an almost limitless demand for electricity.
The NRC exercises influence through engineering evidence, scientific review and independent regulation.
Only one of those institutions exists solely to protect the public.
That imbalance is important.
Microsoft, Amazon, Google and Meta are investing tens of billions of dollars in nuclear-powered AI infrastructure while collectively spending tens of millions of dollars each year lobbying Washington on issues including energy policy, permitting reform and advanced nuclear deployment.
Their commercial incentives are entirely rational. Artificial intelligence requires extraordinary quantities of electricity. Reliable nuclear generation has therefore become a strategic corporate asset.
The administration also has understandable objectives. It wants America to lead the global AI race, expand domestic energy production and remove barriers to infrastructure development.
None of those ambitions are inherently unreasonable. The danger emerges when political urgency and commercial urgency begin reinforcing one another while the independent regulator becomes the weakest institution in the room.
The sovereignty test
We often think of sovereignty as ownership.
It is also the ability to make independent decisions free from external coercion.
By that definition, the independence of regulators matters every bit as much as the ownership of power stations.
A country may own every reactor within its borders, but if the institutions responsible for regulating those reactors become subordinate to political expediency or commercial influence, a different kind of sovereignty has already been lost.
This is why the recent evolution in the NRC’s thinking on artificial intelligence is simultaneously encouraging and deeply troubling.
Technically, the commission appears finally to recognize that adaptive AI systems present regulatory questions that deterministic software never did. Questions of validation. Explainability. Cybersecurity. Model assurance. Those are precisely the questions a responsible nuclear regulator should be asking.
Politically, however, its freedom to ask those questions appears to be shrinking.
That is the paradox.
America unquestionably needs more electricity. It will almost certainly need more nuclear power. It undoubtedly wants to lead the world in artificial intelligence.
But none of those ambitions eliminate the need for an independent regulator capable of asking inconvenient questions when everyone else wants quick answers.
And once a regulator loses the sovereign ability to say “not yet,” it has already ceased to be a regulator. It has become an administrator.
And once a nation’s nuclear regulator becomes an administrator rather than an independent guardian of the public interest, the question is no longer who owns the reactors.
It is who governs them.
Related Fierce Network articles about nuclear power
Op-Ed: Unregulated nuclear AI is here | Fierce Network
Meta goes all-in on nuclear power with 6.6 GW plans
Stephen M. Saunders MBE is a communications analyst and USPTO-registered inventor examining how digital infrastructure — 5G, cloud, and AI — is reshaping industry, power and society, as well as underpinning the emerging, ubiquitous global digital economy. As anchor of FNTV and a longtime industry insider, he focuses less on growth narratives and more on execution, risk and how hyperscale technology is distorting markets, governance and society at scale.
Opinion pieces from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.