A live, sourced map of how the world's biggest AI companies finance each other through equity, compute and venture deals. Hover or click any node to follow the cash.
Nvidia takes equity in an AI lab. The lab uses that money, plus compute deals with Oracle, AMD, CoreWeave and Broadcom, to buy GPUs. Most come from Nvidia. Nvidia's revenue rises (Q1 FY27 hit $81.6B, up 85% year-on-year), its share price rises, and it uses the proceeds to take equity in more AI companies, who buy more GPUs.
The same loop runs through the hyperscalers. Microsoft owns 27% of OpenAI; OpenAI just committed $250B back to Azure. Amazon has now put $83B+ into Anthropic and OpenAI combined; between them, they've committed $238B+ of compute back to AWS ($138B from OpenAI across two tranches, $100B+ from Anthropic). Google has put $40B+ into Anthropic; Anthropic will deploy a million Google TPUs. In July 2026 AMD joined in, putting up to $5B into Anthropic in exchange for a 2GW chip order. Most of these deals are still letters of intent.
July was the month the loop stopped being a metaphor. Nvidia signed or opened talks on more than $750B of deals in a single week: a $500B+ letter of intent with SK Group, $5B into Ilya Sutskever's revenue-free Safe Superintelligence in return for it buying ten times more Nvidia compute, and — most striking — a proposal to guarantee up to $250B of OpenAI's data centre lease payments plus $350B of its chip purchases. A supplier underwriting its customer's ability to buy from it is no longer an analogy for vendor financing. It is vendor financing.
Then the bill arrived. Between 24 and 29 July more than $1 trillion of market value came off global chip and AI-hardware stocks, and the credit market moved faster than the equity one: CoreWeave's five-year default swaps hit roughly 855 basis points — a coin-flip chance of default on standard models — and its lenders started demanding covenants. Alphabet posted its first negative free cash flow as a public company. SoftBank's own default protection repriced, with $30B due before the year is out and lenders refusing to lend against its OpenAI stake at all. And then, within a week, almost all of the equity drawdown was bought back: Nvidia rose 10.5% in five sessions and closed above $5 trillion again. The market now reprices the risk and unwinds the repricing inside the same fortnight.
The financing keeps finding thinner ground. On 4 August, Anthropic signed a six-year, $10B compute agreement with Volta — a company seven months old, backed by Nvidia, whose ability to deliver rests on $1.3B of J.P. Morgan letters of credit and a sixteen-year lease on a former bitcoin-mining site in Norway. Nvidia, meanwhile, disclosed that it owns 9.3% of Nebius, the neocloud renting Nvidia chips to Microsoft and Meta for up to $46B. There is now a bank guarantee, a chip vendor's equity stake and a hyperscaler's rent cheque stacked on top of the same GPUs.
Hover any line to see the deal. Click any company to see every flow it's in. Use the filters to isolate the equity flows from the compute flows.
Bubble size is roughly proportional to current valuation or market cap. Many headline figures are letters of intent, not signed contracts; status is shown in each deal's detail. The Nvidia–OpenAI $100B was, per Nvidia's own CFO, still a non-binding letter of intent as of December 2025, and the $600B of Nvidia financing reported on 26 July 2026 is at the talks stage — the Wall Street Journal notes it could fall apart.
Fortune ran the numbers in late April 2026. Once you back out the paper gains on Anthropic equity from Google and Amazon's earnings reports, the underlying AI business is meaningfully less profitable than it appears.
Bloomberg has been running a permanent graphic called "AI Circular Deals" since early 2026. The IMF returned to the theme in its July 2026 World Economic Outlook Update, warning that frothy valuations in AI-exporting economies could correct sharply. Earnings reports now show investors drawing an explicit distinction between revenue from third parties and revenue from companies the supplier itself has just invested in.
On 12 June 2026 SpaceX began trading on Nasdaq (SPCX) at a $1.77T valuation, raising $75B in the largest IPO in history. Its S-1 disclosed that Anthropic pays $1.25B a month and Google $920M a month for Colossus compute access; Reflection AI — whose largest investor is Nvidia — added $150M a month from 1 July. That is roughly $2.3B a month, about $28B a year, flowing from three AI labs into a newly public rocket company. By late July the stock was trading below its offer price. SpaceX's own Q2 capital spending came in at $18.4B, against $2.8B a year earlier.
Since the end of July the argument has moved from the equity market to the credit market, where it is harder to talk up. Moody's has warned that AI spending threatens the credit quality of Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave. CoreWeave's default swaps briefly implied a coin flip. Anthropic's newest compute contract is underwritten by a bank letter of credit rather than a balance sheet. Equity investors can reprice enthusiasm in a week; lenders write covenants that last the length of the loan.
"The idea that it is circular is — it's ridiculous."
Jensen Huang, Nvidia CEO, responding to the vendor-financing charge — quoted again as Bloomberg counted $750B of new Nvidia deals, 27 July 2026
"Between us, we're going to do half a trillion dollars' worth of business."
Jensen Huang on the SK Group letters of intent, 25 July 2026
"The AI ecosystem requires $2 trillion in annual revenue by 2030 to justify current infrastructure spending. The trajectory falls roughly $800 billion short."
Bain & Company, Global Technology Report, September 2025 — the next edition is due September 2026
Frothy equity valuations — particularly in AI-exporting economies and markets concentrated in technology firms — could correct sharply, amplified by higher risk sensitivity among AI-exposed investors.
Paraphrasing the IMF's July 2026 World Economic Outlook Update, presented by chief economist Pierre-Olivier Gourinchas. The Fund's "AI disappoints" scenario models a ~1.2% hit to global output.
Historical scores are applied retrospectively by hand, using the same six indicators. They are judgements about the record, not measurements — each carries its basis. The telecom marker is the closest structural analogue to what this map shows: Lucent, Nortel and Motorola lent their own customers the money to buy their equipment, booked the sales as revenue, then wrote the loans off.
What this is. A measure of how closely the current picture matches the characteristics common to historical asset bubbles. It is not a forecast, it says nothing about timing, and it is not investment advice. Half the weight is computed from the 60 deals tracked on this page; half is hand-set and sourced.
What it is not. These 60 deals were selected because they are circular — that is the point of the map. The circularity figure describes this deal set, not the market as a whole, and the denominator is always shown. An index where every needle pointed the same way would be a press release; market behaviour currently points the other way, and says so.
In rough chronological order. Status tags show whether something is signed, completed, a letter of intent, or paused. Click any source to read the original reporting.