The AI Boom Is Starting to Fund Itself. That Should Worry You.

AMD is reported to be investing up to $5 billion in Anthropic. Read the fine print and the deal folds back on itself: Anthropic will spend tens of billions buying AMD's AI servers starting in 2027. The chipmaker is funding the customer that will buy its chips.

This is not a one-off. It is quietly becoming the operating system of the AI economy, and it should change how you read every "record investment" headline this year.

## The template

Nvidia wrote the playbook. It took a stake in OpenAI, and OpenAI buys Nvidia GPUs by the warehouse. Now AMD is running the same loop with Anthropic (the deal terms are here). At the same time, Meta is in talks to rent Anthropic up to $10 billion of compute, another turn of the same wheel where rivals bankroll each other's capacity (we broke down what that repricing means).

On paper, everyone wins. The chipmaker locks in future demand and books the check as a strategic investment. The model lab gets capital and guaranteed supply. Valuations on both sides climb. The press writes it up as another sign the AI market is on fire.

## Why the circle is worth watching

Strip the labels off and a chunk of what the market calls "AI investment" is a supplier financing its own revenue. When AMD invests $5 billion and Anthropic spends it on AMD chips, the "investment" and the "compute demand" are largely the same dollars, counted twice on the way around.

That is not automatically bad. Vendor financing helped build the telecom networks and the early internet. It turns dangerous only when the end demand fails to show up, because the loop holds together only if someone outside the circle, real enterprises and real consumers, eventually pays for all that inference. The 2000 telecom bust was, at its core, vendor financing that outran actual usage. Fiber got laid, gear got shipped, and the traffic to justify it arrived years late.

The AI version has one advantage the telecom bust did not: demand for inference is real and growing today. Moonshot ran out of GPUs 48 hours after opening signups. People clearly want the product. The open question is whether they will pay enough, and soon enough, to make tens of billions in circular capacity commitments look like foresight rather than an accounting loop.

## What to do with this

Two things you can act on this week.

First, when you see an AI "investment" headline, check who sits on both sides of the deal. If the investor is also the supplier being paid, discount the number in your head. Our deal tracker lists the source and target company on every deal, which makes the circular ones easy to spot once you look for them.

Second, watch the one metric the circle cannot fake: outside revenue. Are these labs signing paying enterprise customers at the pace their compute commitments imply? If capacity promises keep growing while external revenue lags, the circle is getting thinner, and the risk is moving onto the balance sheets of the chipmakers underwriting it.

The AI boom is real. So is the temptation to make it look bigger than it is by passing the same dollar around the room. Learn to tell the two apart, and most of this year's headlines get a lot easier to price.