The Loop Has a History

The circular financing driving the AI boom looks unprecedented. It isn't. Two old frameworks explain most of it, and the parts they don't explain are the parts worth worrying about.

Here's the arrangement everyone is squinting at. Hyperscalers pay Nvidia enormous sums for chips. Nvidia takes some of that money and invests it back into those same customers, and into new ones, so they can buy more chips. It funds OpenAI, which spends on compute that runs on Nvidia. It backs neoclouds like CoreWeave, whose whole business is buying Nvidia hardware. It anchors a half-trillion-dollar financing platform to keep the capital flowing. Money leaves the center, comes back as demand, and leaves again.

The two reflexes about this are that it's unprecedented and that it's a house of cards. Both are incomplete. It isn't unprecedented, and the history tells you which parts are a house of cards and which parts are something more durable and more concerning. The loop has a shape, and the shape has scholarship.

The cycle: we're in the installation phase

Start with the economist who mapped this pattern before AI existed. Carlota Perez, in Technological Revolutions and Financial Capital, argued that every major technological revolution of the last 250 years has run the same course. It opens with an installation phase, in which financial capital, not productive capital, pours into the new technology and builds its infrastructure ahead of demand. That phase culminates in a frenzy and a bubble. Then comes a turning-point crisis, and only after it a deployment phase, when the technology diffuses through the whole economy and the productivity finally shows up.

Her canonical cases are canals and railroads. Both were built out by capital markets far faster than the traffic justified, in a speculative frenzy that made and destroyed fortunes, and both left behind a network that remade the economy once the dust settled. That is the part worth holding before you dismiss the AI buildout as pure mania. In Perez's telling, the frenzy is not a flaw in the system. It is the mechanism. Financial capital's job in the installation phase is exactly this: to overbuild the rails, wastefully and speculatively, so the deployment phase has something to run on.

By that map, the circular financing is not an anomaly. It is what the installation phase of a capital-intensive revolution looks like when the productive returns haven't arrived yet and financial capital is doing the building. The railroads had their version too, funded by bonds sold to investors on the promise of traffic that didn't yet exist.

So far, so reassuring. This is where most Perez-flavored takes stop, and it's where the more interesting question starts. Because Perez explains the cycle. She does not explain the control.

The turn: the loop is a money trust

For the control, you have to leave the economists and go to 1912.

That year, a congressional committee led by Arsène Pujo investigated what reformers called the Money Trust. Its target was J. P. Morgan and the mechanism he used to run a web of supposedly independent companies from the center: the interlocking directorate. Morgan's partners and allies sat on the boards of banks, railroads, and industrial firms all at once, so a small circle could coordinate decisions across businesses that looked, on paper, like competitors. Louis Brandeis wrote the era's indictment of it, Other People's Money, the following year. The lesson of both was simple: control doesn't require ownership if you sit at the center of everyone's capital.

Now look at Nvidia's stakes again. They stop looking like a bubble mechanism and start looking like a control mechanism. An equity position in OpenAI. Backing for CoreWeave and other neoclouds. A hand in the financing that lets a dozen "independent" AI companies buy the one thing they all need. These aren't only bets on demand. They are the strands of a web, and Nvidia sits at the center of it, the way Morgan sat at the center of his. The customers are bound to the keystone not just by the product they depend on, but by the capital that funds them.

This is the part Perez's cycle doesn't capture. The installation-phase frenzy is impersonal, a market condition. The money trust is not. It's a structure of power.

Why it looks like it has no rival

The reason this arrangement faces so little effective competition is also, in part, a matter of scholarship.

Nvidia is what strategy researchers call a keystone firm, the orchestrator of a business ecosystem, and its CUDA software is the platform that holds the ecosystem in place. That is a well-studied kind of dominance, and it is durable precisely because it isn't a simple monopoly on a product. It's control of the layer everyone else builds on.

And the tools that once checked this kind of power don't fit it. Since the 1970s, American antitrust has run on a consumer-welfare standard that mostly waves through vertical integration, on the theory that a company integrated up and down a stack, but not across a single market, still competes at each layer and so won't raise prices. That theory struggles badly with a firm that supplies the chips, writes the software, sells the systems, and funds the customers. The legal machinery that broke Standard Oil and dissolved the money trust was built for horizontal monopolies and cartels. It has no easy grip on a keystone that owns a stack and banks its own buyers. That mismatch is a live topic in antitrust scholarship right now, and it is a large part of why the investigations into Nvidia keep stalling.

There's one more echo worth naming. In the panic of 1907, Morgan, a private citizen, personally organized a rescue of the banking system because no public institution had the authority or the resources to do it. The spectacle of a single private actor holding up the national economy is what finally produced the Federal Reserve. A firm whose financing becomes load-bearing for an entire sector invites the same question: what happens to everyone else if the keystone stumbles, and who is positioned to catch it?

The two things the railroads never had

If the story were only "installation-phase frenzy plus money trust," it would be a clean historical rerun. It isn't, and the differences are the sharp edge.

First, the loop is also the financing. The railroads raised money from outside investors to build ahead of demand. Nvidia is doing something the railroads couldn't: it is funding its own demand. The buyer's ability to buy is, in part, created by the seller. That folds the installation-phase capital and the money-trust control into a single move, and it means the demand signal the whole market prices off is partly manufactured by the company that benefits from it.

Second, it's geopolitical. Morgan's web was national. Nvidia's reaches across borders, into deals with governments framed as national AI strategy. The last Gilded Age ended when a domestic political system decided it had seen enough and built the countervailing institutions: the Fed, the antitrust cases, the Progressive Era. This buildout is wrapped in an international race, and a race is the one frame under which a government is least likely to move against its own champion.

Does the reset come?

That is the open question, and Perez, usefully, is the one who frames it. In her model the crisis isn't the end. It's the hinge. The turning point is where institutions catch up to the technology, rein in the financial excess, and clear the way for deployment and the golden age that follows. The last time we ran this cycle, the reset arrived: the money trust was investigated and broken, the central bank was created, and the rails, freed of the worst of the financial engineering, did go on to remake the economy.

So the question for this cycle isn't whether the buildout is real. Like the railroads, it probably is, and it will probably leave behind infrastructure that matters. The question is whether the reset still comes when the keystone funds its own demand and shelters under a national-security banner that makes any move against it look like unilateral disarmament.

The frenzy funded the rails last time. Then the country took the rails back from the men who built them. It's worth asking whether we've kept the part of the story that does that.