Nvidia's $10 Billion Anchor Bet Is Really About Who Owns AI

Two different investments are wearing the same headline this week. Nvidia is in talks to put up to $10 billion into Anthropic's initial public offering — the one seeking to raise up to $100 billion at a valuation around $2 trillion, which would make it the largest IPO in history.

One of those investments deserves your attention. The other deserves your skepticism. The entire skill is telling them apart, and this headline shows you the seam.

Start with what the $2 trillion price tag is actually resting on. Anthropic is not being priced on profits, or even on this year's revenue. It is being priced on a number that does not exist yet: a forecast that revenue reaches $190 billion to $200 billion in 2028. The reported run rate climbed from about $9 billion at the end of 2025 to roughly $47 billion by May 2026 — extraordinary growth, no argument about that — but the valuation asks you to multiply that by four inside two and a half years, then pay up for it. That is a 2029 accounting discussion being priced on day one.

Look at the comps bankers are using and the bar gets visible. Palantir trades at about 53 times this year's expected revenue; Cloudflare and Space Exploration Technologies around 41 times expected 2026 revenue. When a market wants to justify a while-you-wait number, it reaches for the richest comparable and leans forward. Nothing about that structure is cheap if the 2028 forecast slips.

The part the "biggest IPO ever" banner buries

Here is the tension worth slowing down for. Anthropic's revenue is usage-based: metered per token, repriced constantly, with no long-term contract, and cancellable within weeks. Its costs are the mirror image — fixed, multi-year, contractual. The company carries roughly $1.25 billion a month in fixed commitments through May 2029, about $15 billion a year, near a third of expected 2026 revenue. And the counterparty on that bill is SpaceX, which now owns xAI — a direct competitor.

That is not a detail; it is the definition of the business. Demand can be turned off in a week. The bill cannot.

A former Goldman and JPMorgan banker who studied this structure draws the parallel to the long-haul fiber build-out of 1999 to 2002: fixed multi-year capacity commitments sized against extrapolated demand curves, then unit prices collapsed faster than traffic grew, and the capacity turned into stranded cost. He puts it sharply: a contracted enterprise book does not bend inside nineteen days; a metered service with cancellable demand does. That nineteen-day figure is not random — in June, U.S. export controls pulled Anthropic's two leading models offline for nineteen days, and revenue bent while the fixed cost did not.

So the "biggest IPO ever" is functionally a wager that a business with termless demand can grow fourfold fast enough to bury a fixed bill owed to its competitor. It may well happen. But as a setup it is paying full price for a hoped-for future and swallowing the open tail risk as a free extra. That is not asymmetric; it is the opposite.

The other half of the headline is the scarce asset

Now look at what Nvidia is actually doing, because that is the more revealing of the two investments.

Nvidia already sells Anthropic the shovels — Anthropic runs on Nvidia chips and is adopting a gigawatt of compute on Nvidia's latest systems. But Nvidia is no longer content to just be the supplier. It has invested nearly $50 billion in the frontier AI labs, it is behind hundreds of billions in guarantees tied to data-center financing, and now it is considering a $10 billion anchor stake in the biggest lab's listing. The coverage itself describes the move as a "lock" more than a wager.

This is the scarcity story wearing a headline. Money is the abundant thing in AI — capital is pouring into the model labs from every direction. The genuinely scarce thing is compute: the chips and the power to run them. That is what Nvidia owns. By taking equity in the people who buy it, Nvidia turns a supplier relationship into ownership of every lab at once, without betting on any single one.

The asymmetry sits there. If Anthropic wins, Nvidia's stake appreciates. If Anthropic stumbles, Nvidia still sells Anthropic chips — and OpenAI chips, and everyone else's chips. The individual model labs carry their own bets to their own shareholders. Nvidia is diversified across the whole pool of miners while each miner bets on its own strike.

Scale makes the point concrete. Nvidia generated about $21 billion in free cash flow in a single quarter, and the $10 billion it might anchor is a rounding error against that machine. A sum that would be the defining event of most companies' existence is, for the owner of compute, a modest check. That is the clearest signal yet of where the durable value in this cycle actually lives — and it is not inside the 2028 forecast that carried Anthropic to $2 trillion.

What you do with a headline this big

Separate the two questions it is asking.

If you are drawn to the IPO, you are buying a forecast and a multiple, not a track record — with revenue that has no term, a fixed bill owed to a competitor, and a regulatory tap that has already been shown to turn off in a week. The upside is the fourfold growth; the downside is open and priced in as if it is not there.

If your question is where the real economics of AI compound, this move is strong evidence. The company that owns the scarce input is using its enormous cash generation not to gamble on one lab but to buy a slice of them all, capturing the economics whichever model company wins. When the largest chip seller in the world starts writing checks to its own customers, it is telling you who holds the cards. It sells the picks and shovels — and now owns a share of the miners. Everyone else is betting on which miner strikes gold.