The AI Boom Had a Hidden Banker, and He Just Left

Share

For the past few years we have told ourselves a tidy story about the AI boom. Founders raise money, labs ship models, and whoever builds something people want wins. It is a good story, and like most good stories it leaves out the figure standing in the corner of nearly every frame. This week that figure stepped away, and it is worth stopping to look at what he actually was before the moment passes and the industry moves on to the next thing.

Jensen Huang, who had led Nvidia essentially since its founding, resigned as chief executive this week. The market mostly processes that as a leadership change, a succession event at a very large company. I think that undersells it. Nvidia under Huang was not merely the supplier of the most sought after chips in the world. By the accounting in the New York Times, it held roughly $48 billion in stakes across 229 private companies, a collection that includes OpenAI and Anthropic, the two names at the center of the whole excitement. Strategic investments are normal for a successful chip company. It is a different thing entirely to be, in quiet practice, the largest patron of an entire technological era.

What made the position genuinely unusual is that Nvidia occupied three roles at once. It was the dominant seller of the scarce resource that the whole boom depended on. It was an investor in the companies buying that resource. And at moments of real trouble it behaved like an emergency funder as well. Founders who lived through the Silicon Valley Bank collapse recall a CEO who stepped in personally, prepaying invoices so a struggling startup could keep the lights on. I have a hard time naming another company in business history that played arms dealer, banker, and customer to an entire industry simultaneously, and did all three well enough that no one thought to question whether they should.

A fair amount of that reassurance was personal rather than institutional. People describe a founder who called and emailed other founders in hard times, who treated Nvidia's position as a responsibility instead of just a windfall. One founder put it more plainly than any analyst could: when Nvidia was on the cap table, everyone felt safe. That is not the language of a vendor relationship. It is the language of a person. A board can carry a strategy forward after a leader retires. A specific temperament, the willingness to pick up the phone and pre-pay an invoice for a company that is not even a customer yet, is much harder to pass down.

Here is the strangeness that I keep coming back to. The AI boom sold itself as a story of open markets, pluralism, and healthy competition. Labs competed, founders shopped around, nothing was beholden to anyone. Yet a remarkable share of the ecosystem's underlying financial confidence sat on one relationship with one company led by one person. That is a dependency by any definition, and the dependencies you do not notice are the only kind that ever catch you. The market did not price this one in, because it did not see it as a risk. It saw Nvidia as a supplier whose bills happened to be paid and whose shares happened to back half the term sheets in the industry.

I should complicate my own case here, because the obvious objection is real. Institutions exist precisely to outlive their founders. Nvidia still has every strategic reason to keep its customers alive, since its own growth depends on them buying chips. I would not predict the company will stop supporting the ecosystem, and I certainly would not predict the industry collapses because one leader stepped down. There is real continuity in a company that size, and the machine does not stop when the personality leaves. This is not the end of Nvidia, and it is not the end of AI.

But the more interesting question is who inherits the role if the softer parts of it fade. Already venture investors and founders are bracing for smaller checks and lower valuations, a step change from the era when Nvidia's name on a round felt like a guarantee. The companies likely to compete for Nvidia's former centrality, the large cloud providers and the other chipmakers, are all credible candidates. And yet none of them has the same incentive to fund young, pre-revenue infrastructure at the pace Nvidia did. That model, writing big checks into companies that were little more than chip customers in waiting, is not one a cloud provider's careful spreadsheet culture reproduces by default. The enthusiasm can be encouraged in a boardroom. It is harder to commission.

There is a useful parallel in business history, the way a single generous patron shapes a creative scene until they step away. When the patron withdraws, the money and the taste both leave together, and the survivors discover how much of what felt like their own success ran on somebody else's continuing interest. The AI version is cleaner. Whoever controls the scarce input increasingly controls the rest of the story. That was true of GPUs, and it turned out to be equally true of the capital that let people buy them. The two were the same lever, wielded by the same hand.

What actually worries me is less the loss of one person than what this episode says about the balance sheet underneath all the enthusiasm. A great many AI companies built their plans on assumptions that were never written down. Nvidia would always be there as supplier. Nvidia would always hold their paper. Someone sympathetic would always be reachable at the top when things went wrong. None of that was in a term sheet, and term sheets are the only promises that survive a resignation. Comfortable dependency is still dependency. The industry has now learned, fairly gently, what that particular one felt like.

This also sharpens the picture of concentration that has been building all year. If the patron who spread the money around recedes and the cloud giants and self-funding labs keep absorbing the output, a few very large balance sheets end up paying for and owning even more of the frontier. That is not a collapse, and it may not even be a bad outcome for users. But it is a different competitive landscape than the one founders pitched into, and it is worth seeing clearly while the change is still new rather than after it has quietly become the assumption.

I do not mean to be dramatic about any of this. Leaders leave, companies adjust, and the AI industry is far bigger than any single office, even a very large one. But I think the genuinely useful thing to notice is where the leverage actually sat all along. It was never really in the leaderboard or the model card or the headline benchmark. It was in the ability to supply and fund the people chasing those things at the same time, from the same place. The rest of the industry is only now discovering what its own balance sheet is worth.

It was never quite a vendor relationship. It was a relationship, and relationships turn on the person. Nvidia will keep selling chips for a long time. The quiet part, the part where a lone executive's judgment stood behind an entire industry's confidence, is what no one can schedule a succession for.