AMD invests in the customer buying its chips: Anthropic's IPO puts a price on the promise

By Zendoric · September 6, 2026.

AMD has locked in two commitments with Anthropic that are worth keeping separate. The first, announced in late July: Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 series GPUs (graphics processors), with the first gigawatt coming online in the first half of 2027. The second: AMD committed to investing up to $5 billion in Anthropic's equity. According to AMD's own press release, it is an investment the company "has committed to make in the future," not one it has already made. And according to AMD's quarterly report from early August, that commitment is "subject to certain contingencies" that have not been disclosed, with the outlay expected to be staged through fiscal year 2028. No valuation has been published for that investment.

That contractual architecture — plenty of headline, few explicit conditions — has become the standard among Anthropic's major backers. In April, Alphabet agreed to invest up to $40 billion, $10 billion immediately and the remaining $30 billion contingent on performance milestones. AMD held $1.7 billion in private-company investments at the close of the second quarter; this single commitment, if executed in full, would almost triple that.

The context that makes sense of all this is Anthropic's stock market debut. The only public document so far is a confidential filing submitted in June, but The Information has reported that the company will publish its prospectus after Labor Day (Monday) and could begin trading as soon as late September or early October. According to CNBC, the private market already values Anthropic at close to $1 trillion, with investors projecting a possible listing at around $2 trillion. Revenue explains part of the enthusiasm: Anthropic's annualized revenue run rate topped $30 billion in April and passed $65 billion by the end of July, according to reported figures that have not been filed with the regulator. The company is said to have already raised at least $130 billion in previous rounds, and its IPO (initial public offering) could exceed the $86 billion SpaceX raised in June, the largest offering on record to date. At those valuations, AMD's $5 billion would buy, at most, half a percentage point of Anthropic's equity.

But the figure that really deserves attention is not the size of the stake, it is the direction of the money. AMD is not buying a passive position in a service provider: it is partially financing the very company that committed to buying up to 2 gigawatts' worth of its chips. It is the mirror image of its deal with OpenAI, where it is OpenAI that receives warrants (options) on up to 160 million AMD shares in exchange for deploying its hardware. Here it is AMD that puts up the capital and Anthropic that puts up the chip demand. AMD's data center business doubled year over year to $6.7 billion in the second quarter, 58% of its record $11.5 billion in revenue (up 50% year over year), and the company is guiding to a third quarter of around $13 billion, growth of 41% that, while high, already marks a slowdown at greater scale.

This is exactly the capital circularity pattern we have been flagging in AI infrastructure: when the compute provider is also a shareholder in its customer, chip demand announcements stop being an independent market signal. Part of the spending that fuels AMD's reported growth could ultimately be financed by AMD itself. It is not fraud or accounting window dressing — the undisclosed contingencies could substantially dilute the real commitment — but it is a legitimate reason to discount the enthusiasm with which the market greets every "massive deployment" announcement when there is a check from the provider itself behind it.

This is where Anthropic's IPO matters more to AMD's shareholders than to Anthropic itself. Today AMD has no actual stake to value: it has a conditional promise, with no price, with payment stretching to 2028. A public listing would give that future stake a daily price that would feed straight into its results — AMD closed the second quarter with $425 million in unrealized net gains on listed holdings, most of them in companies that went public during the quarter. But above all, listing forces Anthropic to show quarter by quarter how much real revenue it produces, instead of projections leaked to the press. That discipline is, in our view, the best proof that the compute demand behind this boom is real and not merely circular: if the revenue reported today ($65 billion annualized run rate) withstands the scrutiny of public financial statements, much of the concern about self-financing dissipates. If it does not, the market will know quickly, because a listed company cannot live on figures that are merely reported and never filed with the regulator, as The Motley Fool's own analysis of these numbers points out.

In the short term, this is the uncomfortable part that should not be dressed up: much of the AI infrastructure boom rests on cross-commitments among a handful of providers and customers who are also investors in one another, which concentrates risk and makes it harder to know who is really putting up the money. In the long term, if that compute capacity translates into the models that accelerate medical diagnosis, drug discovery and the cheapening of intelligence as a resource, today's accounting detour will be a footnote. What matters to watch now is duller but more important: whether Anthropic's prospectus, when it is published, spells out customer concentration and provider-investors such as AMD, and whether its revenue figures confirm what so far have been only leaks to the press.