The next nuclear treaty must be about AI — and its energy supply and containment

In 1945, World War II ushered in the nuclear era, where a single decision could lead to planetary collapse — yet, not a single war-related nuclear attack has occurred. That legacy of restraint now faces a new test with AI — an existential threat whose unchecked trajectory risks global systemic failure — amplified by its growing reliance on nuclear energies to fuel its rise through Small Modular Reactors (SMRs), fusion, and fission. 

While AI brings unprecedented challenges, it is not the first time humanity has encountered a technology capable of reshaping the global map, or even ending it. Like nuclear weapons, AI — if unchecked, misaligned, or militarised — could destabilise nations, manipulate infrastructure, and escalate conflicts beyond containment.

This is why nuclear-styled oversight must extend not only to AI’s outputs — compute, modelling, and decision-making—but also to its inputs via nuclear energies, particularly as AI’s energy mix begins to intentionally shift away from fossil fuels, as investments by Google DeepMind and OpenAI’s Sam Altman show. Without emissions and safety governance, AI’s digital expansion could outpace the physical infrastructure sustaining it — particularly as nuclear energy becomes part of its core.

Safely scaling AI

Scaling AI with current and future nuclear energies demands the same level of caution historically applied to missiles and warheads. These technologies should be integrated into AI's energy framework only if safety and emissions are managed simultaneously. Regulating algorithms without considering their carbon footprint, safety, and capacity is akin to overseeing software while ignoring the underlying system. Each of these domains presents a containment challenge that echoes lessons from nuclear governance, requiring a structured, domain-specific response. To govern this nexus effectively, each domain — whether algorithmic, infrastructural, or geopolitical — requires targeted oversight. The following table — a strategic instrument — presents a blueprint laying out the specific containment needs.

A coordinated governance authority

This governance vision demands a global oversight body — which is why an AI-Energy Governance Authority is absolutely needed — to set international standards for containment and ethics. Just as nuclear non-proliferation frameworks helped avert catastrophe, AI must now be governed with similar foresight and coordination. It requires systems-level design, institutional foresight, and enforceable coordination. We need a governance order that includes energy providers, technologists, carbon regulators, grid architects, innovation pioneers — and safety experts — because without them, restraint is incomplete; fragmentation becomes failure.

This conviction stems from my 2023 work as a Director at Velocys, advancing the commercialisation and integration of nuclear energy and Direct Air Capture (DAC) linked to a U.S. Department of Energy (DOE) initiative to produce sustainable fuels (Figure 1). Google’s recent multimillion-dollar investments in DAC and nuclear further reinforce this shift — positioning the technology as critical to both corporate, planetary net-zero goals and AI ambitions. In 2025, United Airlines is investing in DAC for sustainable fuels, confirming that the original diagram is now investable. It establishes a referenceable artefact of how emissions, infrastructure, and innovation can be convened in one system — exactly the kind of convergence now needed for AI, with the caveat of strategic oversight.

Figure 1: U.S. DOE diagram showing nuclear-DAC integration, a blueprint for AI-era energy governance (Public domain image)

The U.S. DOE initiative also aligns with four domains from the above table — AI energy inputs, carbon management, infrastructure, and governance gaps — and introduces a fifth for AI: institutional readiness — ensuring governance is credible, independent, and globally interoperable. For safety to scale in the AI-energy nexus, the right convening power must align these actors into one coordinated system. In AI’s case, governance must be lived, systemic, and not imagined. 

Conclusion: A balanced path forward

The intersection of AI and the nuclear nexus has transcended speculation; it is now a matter of operational urgency, infrastructural development, and geopolitical relevance. This convergence makes restraint not a limitation, but a form of strategic containment. Just as nuclear technologies required multilateral frameworks to avoid planetary collapse, AI demands the same discipline and foresight. When referencing this analogy during a videocast in Pakistan — a nuclear-capable state — it was not for rhetorical effect. It was to reinforce that in regions where nuclear diplomacy is lived — not abstract — the lessons of restraint are neither historical nor optional.

As in the Cold War era, when nuclear capabilities shaped global order, today the AI-energy nexus emerges as a new axis of strategic power. It demands a doctrine linking global security, compute access, energy infrastructure, and emissions accountability — within an integrated containment architecture. This requires a holistically ingrained approach, linked by shared risks, interdependent outcomes, and unified oversight. It will take institutional maturity, international cooperation, and governance built for convergence — not silos. Our future depends on restraint — not just in principle, but in design, doctrine, and discipline. We have the tools. The imperative is no longer abstract.

illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.