727 FinanceIn 2026, the nature of US venture capital money has changed: from votes to directly purchasing the means of production—compute, power, and defense contracts. Three quarters, three kinds of money—labs, exits, and capacity taking turns in relay; 27 billion-dollar financings point to data centers and GPUs; an AI chip company's valuation doubled in 26 days; CVC accounts for a record 87.9% of AI deals; moves by OpenAI, Anthropic, and SpaceX affect the whole picture; the divergence between capital frenzy and tightening interest rates reveals the capacity-ization nature of AI financing and a huge market gap.
Author Kexia, Daily Angel
An AI chip company doubled its valuation in 26 days; a personal agent company saw its valuation triple in 33 days; a data center company raised $3.9 billion in a single round.
This is Silicon Valley in the third quarter of 2026. In the same quarter, the Federal Reserve raised rates for the first time in 14 months, the 10-year Treasury yield closed at 5.27%, and the FTC opened cases against both OpenAI and Anthropic simultaneously.
Capital is celebrating, interest rates are tightening, and giants are postponing IPOs.
This divergence is why we produced this report. We tracked 100 institutions in the core circle of U.S. technology investment (70 VCs/crossover funds and 30 CVCs/sovereign capital) and used public data to reconstruct funding trends for the first three quarters of 2026.
The conclusion first: in 2026 U.S. venture capital, the nature of the money has changed. It is no longer just votes to pick winners, but direct purchases of the means of production—compute, power, wafer capacity, and defense contracts.
In the first three quarters of 2026, global startup funding totaled $679 billion, the highest for the same period in history (according to Crunchbase, 2026-10). U.S. companies took about $515.8 billion of that, up about 44% year over year (according to PitchBook-NVCA Venture Monitor, 2026-10).
But this $679 billion is not all the same kind of money.
**The first quarter was the labs' money.** U.S. venture funding in the quarter was $267 billion, with 5 deals accounting for 73%: OpenAI, Anthropic, xAI, Waymo, and Databricks (according to PitchBook-NVCA, 2026-05). OpenAI completed a $122 billion financing at the end of March, at a post-money valuation of $852 billion—the largest private round in the history of human commerce.
**The second quarter was exit money.** SpaceX completed an $8.6 billion IPO on June 12, the largest ever; global quarterly exit value hit a record $113 billion, including 32 billion-dollar IPOs and 24 billion-dollar M&A deals (according to Crunchbase, 2026-07). Anthropic completed a $65 billion Series H on May 28 at a post-money valuation of $965 billion, overtaking OpenAI to become the highest-valued lab—and on the same day, its IPO preparation was also advancing.
**The third quarter was capacity money.** No more giant rounds appeared, but billion-dollar financings surged to a record 27 in the quarter, accounting for about one-third of the quarter's total (according to Crunchbase, 2026-10). Databricks and Safe Superintelligence (SSI) each took $5 billion, Crusoe took $3.9 billion, Mistral about $3.5 billion, Boring Company $3 billion, and Cognition more than $2 billion. The money did not shrink; it simply spread from "giant rounds" into an "oligarch cluster."
Worth emphasizing is the structural signal of the third quarter: of the 27 billion-dollar financings, more than half pointed to the same category of assets—data centers, GPUs, and power. According to Crunchbase, in Q3 aerospace, robotics, data centers, semiconductors, and energy each had more than $10 billion in funding, and among them the five companies with $3 billion-plus financings were all founded no more than 4 years ago.
Late-stage money is buying the scarcest input of this era—capacity—at a pace of "one round every six months, each bigger than the last."
Jinqiu Ji tracked a sample of 25 VCs, providing a rare microcosmic slice: among the 413 financings in Q3 2026, totaling about $54 billion, the top 10 companies by amount took 39.2%; the full-sample average deal size was $137 million, while the median was only $23.98 million.
The average is 5.7 times the median.
This gap explains the true shape of the 2026 market better than any aggregate number. Under PitchBook's methodology, the shape is even more extreme: in the first half of 2026, of the $412.7 billion in U.S. venture funding, 12 deals of $100 million or more accounted for 87.5%; OpenAI and Anthropic alone accounted for 43% of total global venture funding in the first half.
Why must the money be this large? Because these companies' business models share one thing: capital can be directly converted into measurable capability.
Chip companies need tape-outs, fabs, and rack deliveries; compute companies need to prepay years of power and equipment; defense companies need to front capital for delivery cycles. Etched is the most representative sample of the year: in July Jane Street participated in its financing as an investor, by August it had tested chips, received its first racks, and begun deployment, and it subsequently became the lead investor in a new round, while Etched disclosed more than $1 billion in customer contracts over the same period—investment, validation, and orders reinforcing one another in the same closed loop.
The same logic applies to Crusoe's $3.9 billion Series F: what it is buying is not growth, but the right to lock in energy, racks, and delivery cycles in advance.
In the late-stage AI market, fundraising capability itself is the core competitiveness.
If Jinqiu Fund slices the 25 VCs' Q3 deal-by-deal data by sector, two almost completely different rankings emerge.
By number of projects, AI 2B applications rank first (73 companies)—AI makes more and more niche needs worth founding companies for; by funding amount, AI compute and infrastructure rank first ($10.836 billion, 20.1%), defense second ($5.991 billion), and AI 2B applications third ($5.181 billion). Models had only 7 financings but took $4.878 billion, an average of nearly $700 million per deal.
Three structural threads deserve separate recording.
**The first is the debt-ification of neoclouds.** In 2026, the financing structure of GPU clouds underwent a subtle change: Nebius issued $5.75 billion in convertible bonds, CoreWeave issued $3.7 billion in convertible bonds, Nscale issued $3.36 billion in convertible bonds (led by Third Point), and Lambda took a $926 million Class B term loan. Equity capital, convertible bonds, and project debt are mixed on the same balance sheet—venture capital is directly financing U.S. digital infrastructure through financial engineering. CoreWeave's revenue backlog has reached $99.4 billion, a form that is a hybrid of "utility × leveraged buyout."
**The second is the mainstreaming of defense.** According to Dealroom, defense tech funding for NATO and allied countries so far in 2026 totals $27.1 billion, already 1.9 times all of 2025, with the U.S. accounting for 75%. Anduril completed a $5 billion round in May at a $61 billion valuation; Shield AI completed two $1.5 billion rounds within a year; even Saudi-backed funds and pensions like CPP Investments entered Helsing's $1.8 billion Series E. The venture industry, which once avoided military industry, has made defense a core asset allocation.
**The third is the "dumbbellization" of applications.** Among AI 2B applications, three financings—Cognition, Harvey, and Lovable—totaled $2.95 billion, accounting for 56.9% of disclosed amounts in the sector; the vast majority of other application companies still have single financings in the millions to tens of millions of dollars. The application sector is stratifying itself: a few companies get "quasi-infrastructure" pricing, while the rest return to a conventional pace. Entrepreneurial opportunities are spreading, but superscale opportunities are not increasing in tandem—which is another explanation for large VCs' insistence on top-tier projects.
Based on assets under management and public activity in 2026, we selected 100 core U.S. technology investment institutions and annotated each with a verifiable signature move this year.
The Q3 activity ranking shows a clear fault line. Y Combinator stands alone in a tier of its own with more than 221 seed rounds—the second place, Antler, has only 31—and YC has in effect become the "pre-seed infrastructure" of the entire ecosystem. Insight Partners co-led 18 rounds, a16z led 16 rounds with about $6.5 billion led, the highest in the market, and Sequoia led 10 rounds.
Note that these three kinds of "activity" are not the same kind of money: YC is options from batch incubation, Insight is the density of co-leading, and a16z is the concentration of lead amounts. Platform VCs (a16z, Sequoia, Lightspeed, General Catalyst) have an equally exaggerated advantage on the fundraising side—a16z completed a record fundraising of about $15 billion across six funds in January; PitchBook data show that 73% of institutional fundraising market-wide in Q1 flowed to five top managers, and only 101 new U.S. first-time funds were established in all of 2025, the fewest since 2011.
Big funds are getting bigger, and new funds are getting fewer. This trend will not be unfamiliar to Chinese readers.This may be the most underestimated number of 2026: according to a PitchBook analyst note (2026-07), corporate venture capital (CVC) has already accounted for 87.9% of U.S. AI VC deal value—a historical record; AI also accounts for more than 90% of all CVC deal value.
The playbook falls into three categories. Cloud providers use investment to lock in models and long-term cloud contracts: in Anthropic's $65 billion Series H, Google alone led a $10 billion tranche, and Amazon led a $50 billion tranche. Chip and infrastructure providers use investment to bind the technology ecosystem: Nvidia alone made more than $40 billion in company-level AI equity investments in the first four months of 2026, participated in about 67 VC deals in 2025, and appears on the investor list of almost every neolab and neocloud. Frontier labs themselves have also begun to use capital as a market-entry strategy.
When industrial capital is simultaneously customer, supplier, and shareholder, the boundary of "independent judgment" is being rewritten. Nvidia's $30 billion to OpenAI, $10 billion to Anthropic, and $2 billion to xAI are essentially recycling chip profits into ecosystem control..
Money in the primary market must ultimately find an exit in the secondary market, and the exit story in Q3 2026 is full of tension.
SpaceX completed the largest IPO in history with $8.6 billion raised, and its market value was about $2.02 trillion in late September—but its share price first surged, then fell below the issue price, and then recovered, teaching a lesson to all latecomers. OpenAI postponed its IPO to 2027, and its CFO told all employees, "We will become a public company in 2027." Anthropic pushed its listing window to November, with reported target valuation of about $2 trillion and fundraising of up to $10 billion, and its prospectus even disclosed an "existential risk" warning. On September 30, the FTC opened investigations into both companies at the same time.
The macro coordinates are also worth recording: the Federal Reserve raised rates by 25 basis points in September, the 10-year U.S. Treasury yield closed at 5.27% and the 2-year at 4.87%; inflation rebounded, and oil prices remained high; in September, the CEOs of two labs rarely jointly called for "slowing down" frontier development.
The 27 billion-dollar financings were completed against this backdrop. This shows both the determination of capital and marks the coordinates of risk.
Judgment one: the "capacityization" of money is the most fundamental structural change of 2026. The way these companies are evaluated is shifting from TAM stories to capital expenditure intensity and contract visibility. Whoever's financing can be converted into capacity will have an anchor for valuation; for those who cannot convert it, no matter how appealing the story, it will be exposed when this round of tide goes out.
Judgment two: CVC's 87.9% is not a cyclical phenomenon, but the norm of capital organization in the AI era. Independent VC moats are being forced upward—either earlier (YC's 221 seed deals), deeper (Eclipse's physical industries), or more concentrated (Benchmark-style all-in bets). Institutions in the middle are the most dangerous.
Judgment three: what observers should watch most is not the total, but the gap. The gap of "average/median 5.7x," and the gap that early-stage amount accounts for only 12.5%, show that this is a market extremely unfriendly to latecomers—big money has already bought up key capacity.
For early-stage technology investment in China, the window of opportunity is not in replicating huge amounts, but in what huge capital cannot buy: engineer density, supply chain depth, and application scenarios systematically missed by the U.S. market's pricing mechanism..