Why Nvidia Wants to Anchor Anthropic's Record IPO

Nvidia is said to be weighing an investment of up to $10 billion in Anthropic's initial public offering — backing a raise of as much as $100 billion that would value the Claude maker at roughly $2 trillion and make it the largest IPO on record. The numbers are so large that the natural first question is which investor would buy such a deal. The more revealing question is why the world's biggest chipmaker would want to.

Nvidia is not a normal IPO buyer. It is Anthropic's supplier. The two agreed in November 2025 that Nvidia would invest up to $10 billion in the startup as part of a partnership in which Anthropic committed to buy $30 billion of Nvidia compute. A reported second $10 billion check, this time into the offering itself, would be a vendor buying equity in its own customer — and a customer, at that, who is trying to spend its way into owning a competing supplier.

Watch who else is standing at the door. In July, AMD agreed to invest up to $5 billion in Anthropic in a deal under which Anthropic would deploy up to two gigawatts of AMD's next-generation chips starting in 2027. Anthropic is deliberately spreading its capex across the two largest GPU makers. That is the game Nvidia is playing for: not the equity return on a $10 billion stake, but a claim on the multi-hundred-billion-dollar compute flow of the largest and fastest-scaling buyer of AI infrastructure. Anchoring an IPO is a cheap way to keep that flow anchored to you.

The check is small against Nvidia's own scale — fiscal 2026 revenue came to $215.9 billion — which is exactly the point. This is relationship capital, spent to protect the $30 billion purchase commitment and to blunt a rival that is trying to convert Anthropic into a beachhead against Nvidia's near-monopoly on frontier training chips.

That is the business story. The investment story for anyone watching the IPO is different, and harder to see through the record-setting headline.

A $2 trillion valuation against the roughly $47 billion annualized revenue Anthropic reported in May works out to about 40 times run-rate revenue — a multiple that assumes the growth keeps compounding. But the more structural problem is on the other side of the ledger. As a former Wall Street executive now running a private-equity firm argues in one of the sharpest pieces of pre-IPO analysis, Anthropic's cost base is locked in while its revenue is not. The company carries a fixed compute commitment of about $1.25 billion per month into 2029 — close to $15 billion a year, roughly 30 percent of expected 2026 revenue — while most of its revenue is usage-based, metered, and cancellable with little notice. When U.S. export controls took leading Claude models offline for 19 days in June, revenue growth reportedly slowed; the fixed obligations did not bend at all.

This is the asymmetry worth holding onto. Nvidia's reported check tells you that front-of-line access to Anthropic's compute demand is so valuable that the two biggest chipmakers are paying Anthropic to use them. But it does not tell you that the revenue Anthropic earns from those chips converts into durable profit for the company's own shareholders. The record size of the offering is a function of how much capital the AI build-out demands, not a sign that the economics have been proven.

So the reported $10 billion is best read as two different signals. For Nvidia, it is a small, strategic cost of keeping its largest customer inside its ecosystem. For the retail investor watching from the sidelines, it is a useful reminder of where the durable value in this stack is accumulating — with the vendors selling pickaxes into a capex arms race — and a caution against treating the size of an IPO as proof of the quality of the business underneath it. The deal is reportedly still under discussion and could change, as could the outcome of the roadshow. The question worth following is not whether the offering sets a record, but whether Anthropic's usage-based revenue can carry the fixed costs it has already committed to.