A data film · part two · read part one: The Next Four Billion
Part 1 measured adoption. This is the money side, shown straight: who's funding whom, who's buying from whom, and how often those are the same companies, per public filings and reporting. No advice and no verdicts in this piece. Just the ledger, checked 4 times, most recently Sept 7, 2026. What to make of it is your call.
Where venture capital went · Q1 2026 · $297B total
OpenAI and Anthropic alone took 43% of H1's $510B. Seed deal count fell 30% year over year while seed dollars rose 31%. Bigger checks, fewer founders.
I stopped arguing about whether AI investment is concentrated and started pulling the wires to see where each dollar actually goes. The short version: the money moves in circles. The investor is the vendor. The customer is the portfolio company. The revenue and the funding are often the same dollars at different points on the loop. Here's the machine with the panels off, every number checked against 2 or more independent sources, and the ones that couldn't be are labeled.
Exhibit A · the same dollars, both directions
And the round itself closes the circle: OpenAI's $122B raise at $852B in March 2026 included $50B from Amazon ($35B contingent on an IPO or AGI milestone) and $30B from Nvidia. The company's 2 largest classes of supplier are now also among its largest shareholders. Bloomberg's own headline for its deal-map graphic: how Microsoft, OpenAI and Nvidia keep paying each other.
Exhibit B · commitments vs revenue, one scale, no log tricks
OpenAI's committed compute spend by vendor, 2025-2035 · ~$1.15T itemized, $1.4T total reported
Same scale, on purpose. 2025 operating loss: $20.9B. Run rate: $2B a month in March 2026, ~$3.3B a month (~$40B annualized) by August, with 900M+ weekly users approaching 1B. Fast growth, and the commitments still assume it compounds for a decade.
Exhibit C · who's paying for the build
For scale: Goldman puts global AI investment above $1T in 2026, and just 2 companies, Nvidia and Micron, account for a third of all S&P 500 earnings growth this year. Nvidia is the first company to touch a $5.5T market cap. AI capex is running near 1.8% of US GDP in 2026 per Goldman, and the analyses I could verify credit it with roughly 40 to 45% of this year's US GDP growth, with some estimates running higher.
Exhibit D · the private market, priced
| Company | Valuation | As of | Revenue basis | Rough multiple |
|---|---|---|---|---|
| Anthropic | $965B | May 2026 | $47B ARR (reported) | ~21x |
| OpenAI | $852B | Mar 2026 | ~$40B run rate (Aug 2026) | ~21x |
| xAI (into SpaceX) | $250B | Feb 2026 | $3.2B (2025) | ~78x |
| CoreWeave (public) | $104B backlog | Jun 2026 | $5.1B (2025) | backlog = 20x trailing rev |
Exhibit E · the pipeline beneath it
Read alone, 250 unicorns in 8 months looks like a thriving competitive field. The 3 facts above sit next to it: much of the capital minting them comes from the same concentrated pool, the entry rung beneath them narrowed 30%, and the largest completed exits went to the largest platforms. Call that a healthy pipeline or call it a farm system. The numbers are identical either way; the reading is yours.
Exhibit F · 9 months of consolidation
That's the machine,
measured.
Investors who are vendors. Customers who are portfolio companies. Commitments 100x revenue. A third of the index's earnings growth from 2 tickers. 250 unicorns minted into a funnel whose entry rung narrowed 30% and whose largest exits went to the largest platforms, with the open-model hub's sale to a chipmaker now agreed. I'm not telling you what to do about any of it yet. That's the next piece. This one just puts the wiring diagram on the table.
Method. Every headline figure checked against 2 or more independent sources; items resting on one analysis are labeled (the WSJ off-balance-sheet figure, the Anthropic ARR figure, the GDP-growth range). The Nvidia $100B OpenAI commitment remained a letter of intent as of late 2025 and is labeled where shown. Q1 2026 concentration is 63.3% precisely ($188B of $297B, per Crunchbase and TechCrunch). Multiples are rough: private revenue disclosure is partial and run rates move monthly (OpenAI's grew from $2B to ~$3.3B a month between March and August 2026, which is why its multiple compressed from ~35x at close to ~21x on current run rate). Anthropic totals: Amazon ~$33B committed with $20B contingent; Google ~$40B with $30B contingent. Figures triple-checked; the third pass was adversarial, hunting for contradictions. Single-source items carried as attributed estimates, not facts: the UBS revenue-share estimate and the WSJ off-balance-sheet analysis. The Nvidia-CoreWeave $6.3B capacity backstop (Sept 2025) and the ~$20B Groq asset purchase (Dec 2025) are now multi-source. Nvidia customer-concentration figures reflect the latest 10-Q (Aug 2026). Deal statuses stated as of Sept 5, 2026: the Hugging Face acquisition is agreed, not closed; the Nvidia $100B OpenAI commitment is a letter of intent. Words like "loop" and "farm system" are interpretive labels for publicly reported deal structures; every underlying figure is attributed, and no statement here asserts intent by any company. Nothing here is investment advice; this is a description of published deal structures.
Sources: Bloomberg on OpenAI's $122B round · Bloomberg circular-deals graphic · CNBC on Nvidia/Hugging Face · CNBC on SpaceX/xAI · Goldman via Seeking Alpha on Nvidia+Micron · CoreWeave Q2 2026 results · Goldman on $1T global AI investment · Crunchbase unicorn board · Tunguz on OpenAI's compute commitments · Crunchbase H1 2026 funding · Part one: The Next Four Billion