The structural fact of the month is that Nvidia appears on three different sides of the AI capital market inside four weeks, and only one of those appearances is as a supplier.
It led Safe Superintelligence's $5 billion round, the largest US round of the last week of July, into a company with no product. It joined the investor list on Together AI's $800 million Series C, announced 1 July at an $8.3 billion valuation, alongside lead investor Aramco Ventures. And it is reported by the Wall Street Journal, and covered by CNBC on 27 July, to be in talks to guarantee roughly $250 billion of OpenAI financing for a 10-gigawatt data centre that SoftBank's SB Energy is developing in southern Ohio, with a separate discussion of up to $350 billion for the chips inside it. Reuters could not independently verify the report. These are negotiations, not commitments, and nothing has been filed.
Treat the $250 billion as reported and unconfirmed. Treat the other two as done deals with named parties. Even on that conservative reading, the largest supplier in the market is now a funder of its own demand at three different scales.
The money in aggregate
PitchBook data reported on 9 July puts US venture funding at $412.7 billion for the first half of 2026, close to 30% above the total for all of 2025. AI companies took $355.9 billion of it, 86 cents in every dollar. Crunchbase, counting globally and on a different method, puts global startup investment at a record $510 billion for the same half. The two datasets are not comparable line for line, and neither should be quoted as the other.
The concentration matters more than the total. Joanna Glasner's 23 July analysis finds 60% of global funding, about $320 billion, went into rounds of $1 billion or more. In the US it was 73% of $290 billion, and two rounds, OpenAI's and Anthropic's, account for more than half of that. US startups closed 23 known billion-dollar-plus rounds by mid-year, matching the whole of 2025 with five months still to run. Before this year, mega-rounds were a minority of all funding.
Record totals with flat deal counts is not a broad market. It's a narrow one with very large cheques in it.
The rounds that tell you something
- Together AI, $800 million at $8.3 billion, up from $3.3 billion sixteen months earlier. The company reports annual bookings above $1.15 billion and says customers cut inference costs by as much as sixtyfold against closed models. Investors separately committed to build more than 500 megawatts of capacity for it. Open-weight inference is now being financed as infrastructure rather than as a hedge.
- Safe Superintelligence, $5 billion, Nvidia leading, valuation undisclosed. A pre-product research lab at the top of the weekly table.
- Simile, $200 million at a $2 billion post-money, led by Greenoaks, for AI simulation. Eliyan, $145 million at $1 billion, for chip interconnect. Both in the same week as SSI. The barbell is unusually clean: frontier research at one end, physical plumbing at the other, very little in between.
Crunchbase also counted 34 new unicorns in June, ten of them frontier AI labs, which is the supply side of the same phenomenon.
Fund formation
Abu Dhabi's MGX closed Fund I at $49 billion on 1 July, above a reported $45 billion target, with backers across the Gulf, North America, Asia and Europe. It is the largest dedicated AI fund raised, and it already holds positions in OpenAI, Anthropic and xAI across 14 companies. Mubadala and G42 are its founding partners.
One vehicle now carries roughly 12% of a record half-year of US venture funding in committed capital alone. Sovereign money is no longer a co-investor in this market. It's a price-setter.
What private equity did, and did not do
The counter-signal is the more interesting half of the month.
PitchBook reports that US private equity platform buyouts in software have fallen to a decade low: $16.24 billion of deal value in the first five months of 2026, a run rate around a quarter of 2025's record $156 billion. The figures come via syndication rather than PitchBook directly, so treat the precision with a little caution; the direction isn't in doubt.
So at the exact moment venture is putting 86 cents of every dollar into AI, the buyout market has largely stopped paying software multiples. The obvious reading is that sponsors are unwilling to underwrite recurring revenue whose defensibility is now an open question. What they are buying instead is people.
The services wave
Grant Thornton Advisors agreed to acquire CBIZ for $5 billion, announced 29 July. All cash, $55.00 a share, about a 54% premium to the thirty-day volume-weighted average. New Mountain Capital, which led an investment in Grant Thornton in May 2024, is providing additional equity. Expected to close in Q4 2026 subject to CBIZ shareholder and regulatory approval, after which CBIZ delists and its benefits and insurance segment is separated into a standalone New Mountain-backed company. Worth being precise here: the release is about scale in professional services and does not present this as an AI thesis. Anyone telling you it's an AI deal is inferring.
GTCR-backed Experity acquired Exdion Healthcare on 1 July, terms undisclosed. Here the AI thesis is explicit in the release: chart-to-cash automation, coding, billing and compliance, folded into a platform used by close to half of US urgent care clinics. This is the shape to watch. Sponsor owns a platform with distribution, buys the automation, applies it to a back office that already has the customers.
The declared version of the strategy comes from General Catalyst, which has allocated $1.5 billion to AI-enabled roll-ups on the thesis that agents can automate 30 to 70% of workflows in fragmented, labour-intensive services. That range is the sponsor's own claim and no independent verification of it exists. The template was Titan's acquisition of managed service provider RFA in August 2025: build the platform, then buy the book of clients.
What is contested
Whether supplier-funded demand is ordinary vendor financing at a new scale, or a circular arrangement that flatters everyone's numbers. Nvidia's month is the case study, and both readings fit the same facts. Vendor financing is old, legal and often sensible. What is new is the ratio between the guarantee under discussion and the balance sheets involved.
The second contested point is quieter. Crunchbase's 29 July piece argues the AI era belongs to mid-market companies rather than to incumbents or startups, which is close to the opposite of what the concentration data says is actually happening. One of those two is describing the future and the other is describing the present, and they aren't reconcilable this month.
Watching
Whether the Nvidia and OpenAI arrangement ever appears in a filing.
A $250 billion guarantee that stays permanently in newspapers and never in a document is a different fact from one that gets disclosed. The Q3 PitchBook-NVCA Venture Monitor is the other one to read, and the number to look at is deal count rather than deal value. If totals rise again while counts stay flat, this stops being a funding boom and becomes an accounting arrangement between about five companies.