OpenAI’s response to the Department of Energy on AI infrastructure

OpenAI's latest submission to the Department of Energy reads less like a corporate blog post and more like a lobbying brief dressed up in mission language. The company argues that the real bottleneck for AI progress isn't model architecture or training techniques anymore. It's electrons. Data centers need gigawatts of reliable power, chip fabs need years of lead time, and the US permitting system, built for a slower era, is choking both.

The response leans hard on the framing that infrastructure is destiny, borrowing a line usually reserved for empires and railroads. OpenAI's argument is that whichever country can stand up power generation, transmission lines, and semiconductor capacity the fastest will end up hosting the AI systems that matter most. And right now, the company suggests, that's not a guarantee for the United States. Permitting delays for new transmission lines can stretch past a decade. Interconnection queues for new power generation are backed up in most regions. Meanwhile, China has been adding grid capacity and data center sites at a pace American regulators simply can't match under current rules.

OpenAI's ask to DOE isn't subtle: streamline permitting, prioritize AI-relevant projects in federal energy planning, and treat compute infrastructure as a matter of national competitiveness rather than a private-sector problem to be solved by markets alone. There's an implicit request for the government to lean into public-private partnerships, the kind that let companies like OpenAI, Microsoft, and their peers build faster without waiting a decade for a substation upgrade.

What's notable is how openly the letter ties AI capacity to geopolitics. This isn't OpenAI talking about safety testing or model alignment. It's OpenAI talking about steel, turbines, and grid interconnects, the unglamorous stuff that actually determines who trains the next generation of frontier systems. The subtext is that Sam Altman's company would rather see Washington move at Beijing's speed than watch American AI labs get power-constrained while data center proposals in Texas and Virginia sit in regulatory limbo.

Whether DOE acts on any of this is another matter. Energy policy moves on a different clock than AI hype cycles, and infrastructure decisions made now will shape which companies, and which countries, actually have the compute to matter in five years.