For much of last year, the dominant narrative had been that AI would be constrained by the amount of energy that was available to power it, but it now seems to be shifting towards much more practical concerns.
Eric Schmidt, who ran Google from 2001 to 2011 and remains one of the most closely watched voices in technology, has a different theory about what will ultimately determine who wins the AI race — and it has nothing to do with megawatts. “The real limit to AI is not energy,” Schmidt said. “It’s actually cash.”
His argument is blunt and arithmetical. At roughly $50 billion per gigawatt of compute capacity, building out 10 gigawatts — a plausible requirement for frontier AI development — would cost approximately half a trillion dollars. “How many companies, countries, and so forth can hand an industry a trillion dollars of capital?” Schmidt asked. “Very, very few.”
He identified China as one of the few actors capable of mobilising capital at that scale. “The Chinese could certainly do it,” he said, adding that he wasn’t sure whether they were doing so but intended to find out. For the United States, Schmidt’s answer lies not in government spending but in the structure of its financial markets. “The brilliance of the American capital market allows us to borrow that kind of money,” he said. It is a structural advantage, in his view — one that makes trillion-dollar AI bets financeable in ways that are simply not available elsewhere.
Europe, by contrast, comes up short. “The Europeans can’t do this,” Schmidt said plainly, acknowledging that this is “something they’re sort of sore about.” In his framing, the inability to raise and deploy capital at the required scale is not a political or regulatory problem — it is a deeper structural one, and one with real consequences for where AI leadership will end up. “But this is good for America,” he added.
Schmidt’s reframing is significant — and the numbers bear it out. Big Tech is projected to spend approximately $655 billion on AI infrastructure in 2026, a figure that dwarfs anything Europe’s fragmented capital markets could realistically assemble. Amazon alone has committed $200 billion for the year, with Google close behind at $180 billion. These are not just corporate bets — they are, effectively, sovereign-scale investments being made by private actors, something that is only possible because of the depth and flexibility of American capital markets that Schmidt describes.
The geopolitical dimension of his argument is already playing out. China has surpassed the US as the working location of lead authors at NeurIPS, the world’s premier machine learning conference, for the first time — a signal that its AI ambitions are increasingly backed by domestic talent and infrastructure, not just government rhetoric. Chinese open-source models have also overtaken their American counterparts in global downloads, and 80% of startups evaluated by Andreessen Horowitz are now using Chinese AI models — a striking inversion of the landscape from just two years ago.
Whether capital alone determines the outcome is debatable. Questions are already growing about whether the returns justify the investment. Technology’s contribution to US GDP hit a record 46% in 2025, but much of that was driven by capital expenditure rather than productivity gains flowing through to businesses and consumers. And OpenAI President Greg Brockman’s thesis — that a country’s GDP growth will be driven by the amount of compute it can access — only raises the stakes further for countries that can’t keep up with the spending.
Schmidt’s point is less about who has the best models today and more about who can sustain the investment long enough to matter. On that count, his map of the world — America and China in, Europe out — may prove to be one of the most striking geopolitical observations of the AI era.