US AI Leadership Runs Into a Power Problem

US AI Leadership Runs Into a Power Problem A new Atlantic Council commission report on AI strategy arrives as America's electricity grid emerges as the binding constraint on its technological competition with China. The Atlantic Council's Commission on AI published its roadmap for US leadership on Monday (2026-06-01), framing the contest in familiar terms: intensified geopolitical competition, deepening interdependence, and accelerating technological disruption. Authoritarian states, the report noted, are leveraging AI to consolidate power. What the commission's framing makes harder to ignore is what sits beneath every data centre, every model training run, every inference cluster — electrons, in quantity, at a price.⁵ That matters because the US grid may not be able to deliver. An analysis circulated in energy and infrastructure circles describes an "unprecedented infrastructure crisis" in which America's grid bottleneck could hand AI supremacy to China. The mechanism is straightforward: without reliable, affordable power at scale, data centre buildout slows, model training costs rise, and the economic proposition of deploying AI narrows. The commission's geopolitical ambitions rest on an energy foundation that is already under strain.² China's position is not simply about government spending on chips or algorithms. The Economist has reported that abundant and cheaper energy gives Chinese companies a structural cost advantage when running AI models — and that Beijing is especially focused on applying AI to the physical world through robots, autonomous vehicles, and smart factories, all of which compound the demand for electricity. Where energy is cheap, inference is cheap. Where inference is cheap, adoption is faster.⁴ OpenAI moved into this debate on Monday (2026-05-18), when the company released a 13-page paper titled "Industrial Policy for the Intelligence Age." Sam Altman's proposal was striking in its candour: tax AI's winners, direct the proceeds toward infrastructure. The paper proposed hikes on corporate income alongside a broader policy architecture aimed at keeping people first as automation accelerates. Whether Washington will act on it is another question, but the document at least acknowledged what the commission's report also implies — that the US cannot win an AI race on policy vision alone without committing serious capital to the underlying infrastructure.³ The energy investment picture globally provides little comfort. Sultan Al Jaber, speaking at a live-streamed Atlantic Council event on Wednesday (2026-05-21), warned that current upstream investment of around $400 billion a year barely offsets natural decline rates. He placed global spare crude capacity at roughly 3 million barrels per day and argued it needs to be closer to 5 million bpd to provide adequate buffer. The remarks were among his most extensive public comments since the conflict began. They were aimed at oil markets, but the underlying message — that the world is under-investing in energy infrastructure — applies equally to the power networks AI depends on.¹ Al Jaber's own institution is putting capital into the US regardless. He noted that ADNOC, its international arm XRG, and renewables investor Masdar, of which Al Jaber is chairman, have investments worth $85 billion across 19 US states. That capital flow is partly strategic, partly commercial. Masdar's renewables footprint in America is directly relevant to the data centre power question, even if it was not framed that way in his remarks.¹ The UAE is also pressing ahead on its own energy infrastructure. A pipeline bypassing the Strait of Hormuz was described by Al Jaber on Wednesday (2026-05-21) as almost 50 percent complete, with delivery targeted for 2027. The existing Abu Dhabi Crude Oil Pipeline already carries up to 1.8 million barrels per day. The Hormuz bypass is an oil logistics story, but it illustrates a broader principle the AI commission would do well to absorb: that serious actors don't just publish roadmaps, they build infrastructure.¹ The Atlantic Council's report frames AI leadership as a geopolitical imperative. The energy sector's read on the same problem is more concrete. Data centres need power purchase agreements, grid interconnection queues are running years long in most US regions, and the transmission infrastructure to move renewable generation from where it is built to where it is consumed remains chronically underfunded. A strategy document cannot fix a substation bottleneck. The next signal worth watching is whether the commission's roadmap translates into anything that touches permitting reform, grid investment mandates, or federal power offtake commitments for AI infrastructure. Without that, the gap between US strategic ambition and Chinese energy economics will keep widening regardless of how many summits convene in Washington.²