The next AI race will be won at the credit desk

In August, Nvidia pulled together six of the biggest names in finance— Apollo , BlackRock , Blackstone , Brookfield , Goldman Sachs , and KKR—to set up financing platforms that could mobilize over $500 billion for companies building AI infrastructure. It was the latest move by a company that has gone from making chips to orchestrating the wider playing field in AI—and a sign that growth for the world's most valuable company depends as much on the next big engineering challenge as on whether customers can keep affording its products.

It also points to the wider direction of travel in AI. Even as the world obsesses about model releases, chip capabilities, and data center capacity, the variable that increasingly determines who gets to compete is simply money. The next phase of the AI race between companies and countries will depend in large part on who can pay for it; the action is shifting from frontier labs to credit desks. Whoever can raise capital cheaply and at scale will set the pace, and that race still favors the United States, whose capital markets are deeper than anyone else's. For China and Europe, the road to closing the technological gap runs through the financial one.

Technological revolutions tend to be remembered for their technologies: steam engines, railways, the internet. Just as important is the financial machinery that made them possible: the joint-stock company dispersed the risk of trading voyages; early bond and equity markets financed railroads before they could produce a return; venture capital supplied high-risk money to companies too new to raise it from traditional investors.

AI is reaching its own version of this moment. For half a century, technology was asset-light: software could be written once and distributed to billions at negligible marginal cost. AI is turning technology into a heavy industry. The latest systems require factories, advanced chips, substations, and enormous supplies of electricity. Even the world's richest companies are finding it hard to fund all of this from their income.

The five largest American hyperscalers alone are on track to spend $730 billion on capital expenditure in 2026. A shrinking proportion is coming from their own balance sheets; they are expected to finance up to a third of it with debt.

If the most profitable companies in history have to borrow to stay in the race, what hope is there for everyone else?

Nvidia's deal is one answer. Project finance has long funded infrastructure with high up-front costs, paid down by steady returns over many years: power plants, toll roads, airports. What's new is the attempt to turn compute into an asset class that pension funds and insurers are comfortable owning, the way they already own infrastructure.