The circular AI financing loop that has investors worried
On July 24, AMD announced a $5 billion equity investment in Anthropic alongside a 2-gigawatt deal to supply MI450-series chips. Read those two facts together and a pattern emerges: AMD gives Anthropic capital, and Anthropic uses part of that capital to buy hardware from AMD. Both companies get to announce good news — AMD books a demand commitment, Anthropic books a funding round — without a single external dollar necessarily changing hands beyond AMD's own balance sheet.
This is not an isolated case. It is the dominant financing structure of the current AI infrastructure buildout, and in 2026 it has become large enough, and concentrated enough among a small number of counterparties, that market analysts are drawing direct comparisons to the vendor-financing bubble that inflated telecom valuations in the late 1990s before collapsing in 2001.
How the loop works
The mechanics are simple and, individually, not new. A hardware or cloud vendor takes an equity stake in an AI lab, sometimes paired with a multi-year compute purchase commitment. The lab uses the capital — and often the compute credits directly — to train models and run inference at scale. The vendor records the compute purchase as revenue and cites it as evidence of overwhelming demand for its next-generation hardware. That revenue and demand signal supports the vendor's own valuation, which supports its ability to raise more capital or issue stock, some of which can flow into further investments in AI labs.
What makes this different from ordinary vendor financing — which has existed in enterprise IT for decades — is scale and concentration. A handful of companies are now both major AI lab investors and major AI lab suppliers simultaneously, and the dollar figures involved are large enough to move broader market indices.
The deals piling up in 2026
The AMD-Anthropic deal is one data point among several announced in July alone. On July 25, SK Group and Nvidia signed a $500 billion AI infrastructure and HBM4 memory partnership. The same day, Samsung and Broadcom signed a $200 billion memory and foundry deal tied to AI chip production. Microsoft holds a 27% equity stake in OpenAI and disclosed spending nearly $35 billion on AI infrastructure in a single quarter in late 2025 — infrastructure spending that flows substantially back to its own Azure business and to hardware partners, some of whom also hold stakes in OpenAI-adjacent ventures.
Nvidia's position sits at the center of the pattern. Its roughly $100 billion investment commitment to OpenAI, announced in 2025, is structured alongside compute purchase agreements — meaning a meaningful share of the capital Nvidia commits is expected to return to Nvidia as chip revenue. By October 2025, Nvidia's market capitalization exceeded $5 trillion, making it worth more than the GDP of every country except the United States and China, and AI-related companies accounted for roughly 80% of the gains in the U.S. stock market that year.
Why this worries analysts without being fraud
None of this is illegal, and none of the individual companies are disguising the structure — the deals are announced publicly, often in the same press release. The concern is not deception; it is circularity risk. If a meaningful share of an AI lab's apparent revenue growth and a hardware vendor's apparent demand growth both trace back to the same pool of vendor-originated capital, the market may be double-counting a single flow of money as two separate signals of organic demand.
Ray Dalio has compared current AI investment levels to the dot-com bubble. By late 2025, the top five U.S. companies by market cap accounted for roughly 30% of the S&P 500 — the highest concentration in half a century — and the Shiller price-to-earnings ratio for the U.S. market exceeded 40 for the first time since the dot-com crash. The specific historical parallel analysts invoke is the telecom sector's vendor financing collapse of 2000-2001, when equipment makers like Lucent and Nortel extended billions in financing to telecom startups that used the money to buy the vendors' own equipment. When external demand failed to materialize at the projected scale, both the startups and their vendor-financiers collapsed together.
What to actually watch
The AI circular financing pattern is not proof of an imminent collapse — AI products have real, growing external revenue that telecom bandwidth resellers in 2000 largely did not. But it does mean the headline dollar figures in these deals are a worse indicator of genuine market demand than they appear. Three numbers matter more than deal size: the share of an AI lab's revenue coming from customers with no equity or supply relationship to the lab's own investors; the ratio of debt-financed to equity-financed capital expenditure at the major hardware vendors, since debt cannot be walked back if demand disappoints; and whether compute purchase commitments convert to actual delivered and paid-for hardware on the original schedule, rather than being renegotiated downward as deployment dates approach.
For operators and investors tracking this space, the practical takeaway is to treat announced deal values as a ceiling on demand, not a floor, and to weight any AI lab's growth story by how much of its revenue originates outside its own investor and supplier network.