Published: August 2, 2026 | Category: Venture Capital / Artificial Intelligence

Two labs just soaked up more than 60 cents of every venture dollar in America. This week supplied three reasons to question what LPs are actually exposed to.

OpenAI and Anthropic together absorbed more than 60% of all U.S. venture capital dollars in the first half of 2026, according to PitchBook’s Q2 Venture Monitor — and the question every limited partner is now asking privately is what exactly they are buying access to.

The official answer has been “access, not a stock pick.” After two years of thin distributions, limited partners let their general partners chase OpenAI and Anthropic allocations because missing the round felt riskier than being in it. That logic depended on one assumption holding: that frontier model quality would keep compounding out of reach of everyone else. This week supplied three separate reasons to question it.

Three Signals at Once

1. The capital concentration is now extreme, not just heavy. PitchBook’s Q1 2026 Venture Monitor found the top five deals of the quarter — OpenAI, Anthropic, xAI, Waymo and Databricks — captured roughly 73% of all U.S. venture deal value, and institutional LPs put 91% of new commitments into brand-name, seasoned firms, up from 74% a year earlier. OpenAI ($122 billion raised) and Anthropic ($95.6 billion across two rounds) now account for the majority of all new U.S. startup funding.

2. The “closed frontier” moat looks thinner than it did a month ago. Moonshot AI, a Chinese startup, released Kimi K3 in late July — a 2.8 trillion-parameter open-weight model Moonshot calls “the world’s first open 3T-class model.” Independent benchmark testing found it trails only Anthropic’s and OpenAI’s newest models and beats every other rival, including on head-to-head coding evaluations. Mozilla’s chief technology officer, Raffi Krikorian, said American labs are “clearly worried,” or their own CEOs would not be lobbying Washington against restrictions on open models.

3. Nvidia, Microsoft, Meta and Andreessen Horowitz just told Washington the open-weight model is the future, not a sideshow. On July 24, a coalition led by Nvidia’s Jensen Huang published “Open Weights and American AI Leadership,” arguing that downloadable, inspectable model weights are strategic infrastructure the U.S. should protect rather than restrict. Nvidia, Microsoft, Meta and Andreessen Horowitz signed on immediately, alongside Hugging Face, Mistral and Palantir; OpenAI and Google added their names days later once the omission drew notice. Anthropic did not sign. Its chief executive, Dario Amodei, said the company is not seeking a ban — only mandatory safety testing before release — but the split is telling: three of the biggest commercial backers of the AI build-out are now on record favoring the exact competitive dynamic that would erode the pricing power of the two labs holding 60% of the venture market.

The Market Proxy

Layer onto that a market that has been treating SpaceX’s newly listed stock as a rough proxy for AI-adjacent appetite. SpaceX priced its June 11 IPO at $135 a share, hit an intraday high of $225.64 on June 16, and has since fallen roughly 49% from that peak — closing below its own IPO price for the first time on July 16 and trading around $114 to $116 in the final days of July, pressured by upcoming insider lockup expirations and a large stock-funded acquisition. The comparison is imperfect, but the direction of travel matches the mood: the market’s patience for paying up on “the top of the AI stack, no substitutes” is thinner than it was in the spring.

None of this proves the OpenAI-Anthropic bet was wrong. Frontier labs still command the best researchers, the biggest compute contracts and, for now, the top of most public benchmarks. But “access” was always a euphemism for a valuation argument: that competition could not reach these two companies fast enough to matter. Kimi K3 is one data point, not a verdict. Still, when the chipmaker, the two biggest cloud platforms and the most prominent venture firm backing the labs all put their names on a letter saying the walled garden should not be the only garden — and the market’s own proxy for AI enthusiasm is down almost half from its high — the “it’s just access” pitch needs a better argument than it had in January. The concentration is not the risk by itself. The risk is that LPs priced in a moat that the moat’s own suppliers no longer seem convinced will hold.

Key Figures

MetricFigure
OpenAI + Anthropic share of U.S. VC dollars, H1 2026>60%
Top-5 deals share of Q1 2026 U.S. VC deal value~73%
LP commitments into seasoned firms (Q1 2026)91% (vs. 74% prior year)
OpenAI total raised$122 billion
Anthropic total raised (two rounds)$95.6 billion
Kimi K3 parameters2.8 trillion (open-weight)
SpaceX IPO price (Jun 11)$135/share
SpaceX intraday high (Jun 16)$225.64/share
SpaceX late-July price~$114–116/share (~49% off peak)

Sources