US-China AI Primary Market Capital Flows: H1 2026

Per PitchBook, US venture funding reached $412.7B in H1 2026, nearly 30% above all of 2025, with $355.9B (86%) flowing to AI — 86 cents of every venture dollar. On Crunchbase's count, global venture hit $510B in the half, surpassing 2025's full-year $440B and setting a record for any half-year period, with over 70% going to AI. Concentration is the cycle's defining feature: OpenAI ($122B) and Anthropic ($95.6B) together absorbed 43% of global venture funding, making H1 effectively a two-company market. Late July's marginal change: concentration is not reverting but spilling down the model-compute-energy-channel chain, while new funding talks at DeepSeek and Moonshot push the 'narrative premium' higher still. For primary-market investors, headline totals now matter less than structure — at 86% sector concentration, conventional portfolio diversification logic is breaking down, which frames the analysis in the sections that follow.

OpenAI raised $122B in H1 at an $852B post-money valuation, led by Amazon ($50B committed in February) with Nvidia and SoftBank at $30B each. The IPO cadence shifted: per multiple late-June reports, management leans toward delaying the listing to 2027 — CFO Sarah Friar favors later, citing cash burn, compute commitments and public-reporting burdens, while Altman reportedly refuses to list early below a trillion-dollar price. Anthropic is ahead: it filed a confidential S-1 on June 1, targeting Nasdaq as early as October with Goldman Sachs, JPMorgan and Morgan Stanley underwriting a raise of over $60B; its Series H private valuation is $965B, secondary markets imply $1.05-1.15T, annualized revenue was about $30B as of April, and on July 27 it expanded its enterprise partnership with Cognizant. Regulation entered release cadence for the first time: GPT-5.6 (the Sol/Terra/Luna lineup) rolled out in stages under the pre-release review created by the June 2 executive order, opening broadly only on July 9 — Washington effectively deciding when frontier models reach the public. On hardware, Apple sued OpenAI on July 10 over trade secrets (see the risk section's layered treatment), yet OpenAI still shipped the $230 Codex Micro coding device on July 15.

China's keyword is shifting from 'valuation leap' to 'delivery verification.' DeepSeek closed its first external round of roughly RMB 50B (~$7B) in late May at about $52B post-money; founder Liang Wenfeng personally invested RMB 20B (40% of the round) as the largest single backer, followed by Tencent (RMB 10B), the CATL ecosystem (RMB 5B), and JD/NetEase/IDG (RMB 3B each). On July 15, multiple outlets reported talks with new investors at roughly $71-74B (figures vary), with an IPO anchored to 2027 and possibly earlier. The harder event: DeepSeek V4 went GA on July 20 — a 1.6T-parameter MoE with 1M-token context across the lineup, plus first-of-its-kind peak/off-peak API pricing. Moonshot opened its round at $31.5B pre-money and closed it in late July at $35B post-money with $3.5B raised; with June ARR around $300M, the valuation implies over 100x ARR — crystallizing the 'narrative premium' debate; it reportedly plans a pre-IPO round at up to $50B and may list in Hong Kong this year. Zhipu completed a ~HK$31.4B H-share placement on July 9 (HK$1,588/share, the year's largest single tech placement in Hong Kong), and on July 11 Tang Jie's internal letter launched the 'Touch High' plan. WAIC closed in Shanghai on July 20 with the launch of the World AI Cooperation Organization, the first intergovernmental AI body, headquartered in Shanghai; separately, roughly two-thirds of the world's new robotics unicorns in 2026 are reportedly Chinese.

The late-August news bears out this report's central claim, that concentration is the story of this cycle, and extends it: capital is now crowding not just into the top labs but toward their exits. Reported (Quartz and Fortune, Aug 13): Anthropic's investors expect an October listing at a valuation above $2 trillion, which would make it the largest IPO in history; on Aug 25 Quartz added that the company is pitching IPO investors on a $30 trillion market opportunity. That valuation traces to a single sourcing chain, is unconfirmed by the company, and has no independent verification. China is converging on the same logic. Confirmed (Quartz, Engadget, Aug 13-16): DeepSeek shipped V4-Pro and moved to peak/off-peak API pricing from August 16, with peak rates up as much as 1,100% and off-peak at half price. The tiered pricing this report flagged is now standard practice; the era of buying share with cheap tokens is over. Reported (KrASIA citing IPO Zaozhidao, Aug 11): Moonshot AI set August 27 as the final close of its pre-IPO round at roughly $50 billion pre-money, with a Hong Kong filing planned by September 30. That is some 40% above the $35 billion post-money in this report, and likewise single-sourced. The question at the top of the market has shifted from who gets the money to where it gets out.

This keynote reinforces this report's conclusion on the US side, corrects it on the China side, and supplies the mechanism behind concentration at the funding end.

FutureX's read is a five-row mismatch table. Cycle: the US is late in its mania phase, China early to mid. Funding concentration: three companies took 67.3% on the US side in Q1 2026, while 64.6% of China's AI funding went to early-stage rounds. Exits: no AI company listed in the US in Q1; 155 Chinese companies went public in the first half. Compute: the US is constrained by power and advanced packaging, while domestic AI accelerators have reached a 41% share in China. Openness: the US is mostly closed; China leads on open source, and Chinese open models supply over 45% of OpenRouter traffic.

So "concentration spreads from funding to exits" holds in the US. In China, funding sits in early rounds and the exit window opened first. The two sides cannot run on one clock.

The gap has not closed. AI venture investment in 2025 was $194 billion in the US against $14 billion in China, a 15x difference (PitchBook/NVCA/CAICT). ByteDance's latest secondary trade in April 2026 valued it above $600 billion; Meta's market cap is about $1.47 trillion. On price-to-sales, that is a 55% discount. That is our read.

The keynote also explains why funding concentrates. AI and non-AI companies now live in two different eras. Median revenue growth for listed legacy SaaS is below 20%. The 2021 VC vintage sits at a median TVPI of about 1.0x (Cambridge Associates). Chinese fund-of-funds investment fell 21.84% year on year in H1 2026 (China FOF Research Center). FutureX's judgment: capital markets now price "uses AI" and "is AI" companies on two different durations, and duration risk on non-AI assets is rising. Concentration follows.

Full deck: /reports/open-source-breakthrough (first 5 pages public).

Reported (Bloomberg, September 2, 2026): AI coding company Cognition was in talks to raise about $1 billion at a valuation of about $47 billion.

Confirmed (Bloomberg, Reuters, September 8, 2026): Six days later the company announced a $2 billion Series E at a $48 billion post-money valuation, twice the size first reported; The Information put the figure at more than $2 billion. The round is reported to be led by a16z and Accel among others, with outlets differing on the full list of leads. Investing.com and others report that Cognition's annualized revenue rose from $492 million at its May round to nearly $900 million.

Confirmed (TechCrunch, CNBC, September 8, 2026): Mistral AI closed a €3 billion Series D (about $3.58 billion) led by Samsung Electronics, with the EQT-managed Scaleup Europe Fund and existing investor PSG Equity as co-leads, at a post-money valuation above €21 billion (about $24 billion by CNBC's conversion). The company calls it the largest equity round ever completed by a European technology company; the valuation nearly doubles the €11.7 billion set a year earlier in the ASML-led round. New investors include Advent, BlackRock-managed funds and the Grand Duchy of Luxembourg; a16z, ASML, NVIDIA and Salesforce Ventures returned.

Company disclosure (Mistral announcement and CEO interview with CNBC, September 8, 2026): The company says it operates in 20 countries and supports more than 125 enterprises, including Airbus, ASML and HSBC. Proceeds go to compute capacity, infrastructure, commercial growth and international expansion. CEO Arthur Mensch told CNBC he expects to beat the forecast he gave earlier this year of annual recurring revenue above $1 billion in 2026.

Effect on this report's conclusions: The week reinforces the core finding that the channel is narrowing. Both large rounds went to incumbents with meaningful revenue, and Cognition lifted its raise from $1 billion to $2 billion within six days, a sign that capital supply for top names still exceeds their needs. One correction to the frame: the report casts concentration as a US-China contrast, while Mistral, an open-weight model company, taking Europe's largest round with Samsung, the Luxembourg state and an EU-backed fund entering together shows sovereign AI capital forming a third pool in Europe, and gives Section 3's open-source argument a sample outside the US and China. The scale remains small: €3 billion is about $3.6 billion, a fraction of the tens of billions US frontier labs raise in a single round. No verifiable Chinese funding, filing or valuation change surfaced this week, so Section 3's view on valuation resets stands.

Reported (Bloomberg, September 13, 2026; Yahoo Finance citing Bloomberg, September 15): Anthropic has selected Nasdaq as its listing venue. The offering could raise as much as $100 billion at a valuation discussed at $2 trillion, with Morgan Stanley, Goldman Sachs and JPMorgan leading the book. Second-quarter revenue more than doubled to $11.6 billion. Nvidia is reportedly weighing a contribution of around $10 billion to the offering. Against the "more than $60 billion" raise this report recorded in June, the ceiling has moved up by roughly two-thirds.

Reported (Bloomberg, September 15, 2026; Forbes, September 16): OpenAI is weighing a pre-IPO funding round. Bloomberg puts the valuation above $1.2 trillion; Forbes, citing the same reporting, puts it at $1.5 trillion. Both figures are headline-level; the round size and investors could not be verified for this section. Measured against the March round recorded in this report ($122 billion raised at $852 billion post-money), that is a 41% to 76% step-up within six months.

Reported (Sina Finance and Tencent News citing The Information, September 24, 2026): DeepSeek is targeting RMB 50 billion (about $7.5 billion) for its second round at a pre-money valuation of RMB 500 billion, with closing planned for the end of October, about two months later than the timeline reported in late August. Liang Wenfeng told investors at recent meetings that annualized revenue has reached $1 billion, up from $400 million to $500 million in July; gross margin on the model API business was 82.9% for the first seven months of the year. The company has retained CITIC Securities to prepare a STAR Market listing, plans to file within the year, and targets a 2027 debut.

Effect on this report's conclusions: This reinforces the core judgment that capital has not shrunk while the channel has narrowed. All three new data points from mid-to-late September come from the same top three names and point the same way: Anthropic and OpenAI have lifted their pricing anchors to a $100 billion raise ceiling and a valuation above $1.2 trillion respectively, and DeepSeek is advancing its second round at an RMB 500 billion pre-money valuation. One correction: the previous section wrote that concentration was spreading from fundraising to exits; the exit side is now itself splitting into tiers. Anthropic is set for Nasdaq while DeepSeek prepares for the STAR Market, and the two paths are decoupling in both timing and pricing mechanism. The US-China comparison shifts from "who is valued higher" to "who prices first".

Reported (unpublished draft prospectus obtained by Reuters and also reported by the FT, summarized by Fortune on September 29, 2026): Anthropic's 2025 revenue was $4.6 billion, with an operating loss above $8 billion, a net loss of $42 billion and $20.28 billion in cash and cash equivalents at year-end. Q1 2026 revenue was $4.73 billion. The draft shows plans to spend $518 billion on cloud services and data centers, and a quarter of revenue comes from two clients. According to the FT, more than a third of the document covers business risks, including "existential risks to humanity."

Reported (Bloomberg, October 1, 2026, relayed by The Standard): Anthropic plans to kick off its IPO roadshow on November 9, and people familiar said it still expects to complete the listing by year-end at the latest. It expects to meet prospective investors at its San Francisco headquarters on October 14. The report said some prospective investors put its fair valuation at $1.8 trillion to $2 trillion.

Reported (Bloomberg, September 29, 2026, relayed by TechCrunch): OpenAI is in talks to raise at least $30 billion in a pre-IPO round at a valuation of roughly $1.4 trillion, about 64% above the $852 billion set in March. If completed, the round would serve as a bridge to an IPO; CEO Sam Altman has ruled out a 2026 listing.

Impact on this report's thesis: This strengthens our concentration call ("the river got narrower") and extends it from fundraising to costs. The two leaders now anchor at about $1.4 trillion and at $1.8 trillion to $2 trillion, and Anthropic's draft shows those valuations sit on spending plans in the hundreds of billions and very large book losses, so capital needs are likely to keep pooling at the top. The US-China gap widens: DeepSeek's second round, recorded here on September 25, uses a pre-money valuation of RMB 500 billion (about $70 billion), less than one-twentieth of the bottom of that range. Threshold: if Anthropic ultimately prices below $1 trillion, or its listing slips into 2027, we would downgrade the view that concentration has spread to exits.