Reuters reported on 11 September 2026, citing people familiar with the discussions, that Nvidia is in talks to become an anchor investor in Anthropic's planned initial public offering, "potentially committing as much as $10 billion."

The offering itself would be extraordinary. Anthropic is "seeking to raise as much as $100 billion at a valuation of around $2 trillion", and the IPO is "expected to be completed before the U.S. midterm elections in November."

Anthropic declined to comment. Nvidia did not immediately respond. Reuters notes the plans "remain under negotiation and could change."

We looked at the arithmetic when the filing first surfaced in August. The number has moved since. What is worth examining now is not the size but the shape.

The supplier is becoming the shareholder

Nvidia sells Anthropic the hardware Anthropic's product runs on. If it also becomes an anchor investor in Anthropic's IPO, cash flows from Nvidia into Anthropic, and a large share of it flows back to Nvidia as GPU purchases.

That is not a scandal on its own. Strategic investment by a supplier is ordinary in capital-intensive industries, and Nvidia has an obvious interest in its largest customers being well funded. But it does make two numbers harder to read than they look.

The first is demand. When a chip vendor helps fund the buyers of its chips, revenue growth and investment decisions stop being independent signals. An order backed partly by the seller's own capital tells you less about underlying demand than an order that was not.

The second is valuation. An anchor investor with a commercial relationship is not pricing the asset the way a pure financial investor would, because it is also buying the continuation of a customer.

Anthropic is not only buying Nvidia

The counterweight, from the same Reuters report, is that Anthropic has spread its compute commitments widely:

  • "$30 billion of Microsoft Azure computing capacity powered by Nvidia chips"
  • "more than $100 billion over a decade to Amazon Web Services"
  • a partnership with "Google and Broadcom to add multiple gigawatts of TPU capacity"

That last one matters most to the circularity question. TPUs are not Nvidia hardware. A company committing multiple gigawatts to a competing accelerator is not captured by its GPU supplier, whatever the investment.

This is the pattern, not the exception

Set it beside the rest of what has been reported over the past fortnight. Oracle is building capacity partly funded by prepayments from the AI companies that will use it. A trading firm has been leading rounds and signing multi-year compute contracts. The Pentagon is in talks to lend billions for data-centre components.

In each case the money that pays for AI infrastructure comes from somebody with a direct interest in that infrastructure existing. Very little of it is arm's-length capital taking a view on returns.

That is what makes the sector fast, and it is also what makes it correlated. If demand disappoints, the losses do not land on one balance sheet — they land on several that are already connected to each other.

What to watch

  • Whether Nvidia's participation is disclosed in the prospectus, and at what price relative to other investors.
  • The revenue and loss figures when the filing becomes public. A 2 trillion dollar valuation implies assumptions that a prospectus has to show.
  • Whether the timing holds. Completing before November is an aggressive schedule for an offering this size.
  • What the TPU commitment does to Nvidia's share of Anthropic's compute over time.

What is not established

  • Whether any of it happens. These are talks, reported by one outlet from anonymous sources.
  • The terms of Nvidia's participation, including whether it would be at the IPO price.
  • Anthropic's financials. No public filing yet.
  • Whether 100 billion is the raise or a ceiling, and how much is primary versus secondary.