The era of "picking winners" is over; top VCs have started "making winners." This dashboard tracks two paths: US funds growing PE arms (acquiring, consolidating, building companies from scratch, with frontier labs stepping in directly), and Chinese capital, space and industrial resources nurturing designated players into champions. No market data feed; backfilled by event; every figure notes its source and currency.
Manufacturing winners profits from the execution gap, and the ceiling is capacity. By September, capacity can be measured: capital on the US track exceeds $10B, and China's three model companies took 30% of all funding.
US side: Thrive Holdings raised $2B at a $12B valuation and has acquired 70+ companiescompanies; General Catalyst-incubated Long Lake is taking Amex GBT private for $6.3B; OpenAI and Anthropic have each set up deployment companies, the former with $4B+ in initial capital. China side: in H1, AI took 48.6% of the money but only 22.5% of deals; DeepSeek, StepFun and Moonshot AI combined for CNY 93bn. By early September, all three had reached the IPO door: DeepSeek's second round at ~CNY 500bn pre-money, cumulative raise over CNY 100bn, preparing for the STAR Market; Moonshot AI's Series G at $50B pre-money, confidentially filing with HKEX on September 4; StepFun filed in June. What the two paths share is that winners are designated by capital before being validated by the market.
In the "making winners" chain, success often hinges not on capital but on a few gates nobody has priced: practice rights, corporate practice of medicine bans, signing liability, insurance, single-state licenses, antitrust filings, lenders' views on AI risk, deployment talent, data ownership, model pricing, listing channels and chip deliveries. Each data pull rescans the list: new findings in orange, tightening in red, loosening in green.
Holding-platform list. Three models coexist: buying existing firms (Thrive Holdings acquiring accounting and IT services businesses), building companies (General Catalyst's Creation strategy), and taking public companies private (Long Lake buying Amex GBT, GC buying Janus Henderson). Per Newcomer, capital dedicated to this strategy exceeds $3B; including the two labs' deployment companies, over $10B.
Funding or committed capital, excluding valuation; see table at right for valuations. Ode and Crescendo disclose only valuations and are not charted.
Sorted by most recent event; "acquired" is the number of traditional businesses rolled into the platform.
Spun out of Thrive Capital; OpenAI took a stake in December 2025 and embeds research, product and engineering teams in acquired companies.
A "manufactured" company takes over listed-company-scale assets for the first time; the deal closed on 2026-09-29, and GBTG was delisted from the NYSE the same day.
OpenAI is at once the model vendor, a Thrive Holdings shareholder, and a source of embedded staff; since May all three labs have set up or begun building deployment companies, turning "helping clients transform their business" into a standalone balance sheet. What's been copied is the deployment company; what hasn't yet is equity stakes in portfolio companies.
Per ITjuzi data, China AI financing in H1 2026 reached CNY 307.7bn across 1,203 deals, already exceeding full-year 2025 in half a year. It accounted for 48.6% of total market funding but only 22.5% of deals; the top 30 companies took over CNY 170bn, and the three model companies CNY 93bn, about 30%. 626 very-early-stage deals averaged CNY 73M; 177 late-stage deals averaged CNY 624M. In H2 all three model companies are racing to list: DeepSeek's Jan–Jul revenue was about CNY 475M with a net loss of about CNY 715M, against a CNY 500bn pre-money valuation.
Foundation models take more than half; embodied AI, with 312 deals, is the layer with the most deals.
Four companies vs. the remaining 1,199 deals. DeepSeek's single round of CNY 51bn and the CNY 50bn in the May weekly report reflect different sources.
Hangzhou rose to second on a single DeepSeek round; deal counts in parentheses.
4.3bn at end-2025, 18bn in March, 35bn post-money in the July Series F, 50bn pre-money in the August Series G; on Sep 4 reportedly filed confidentially with HKEX, seeking to raise $3–5bn.
Shenzhen's Delin Capital spent CNY 40M building ~4,000 sq m of residency space, with a CNY 2M cash champion prize taking no equity, equity investments of up to CNY 5M each for up to 15 teams, and 10 weeks of housing and compute included. Incubation moves ahead of product validation, and VCs go from receiving periodic reports to working alongside founders daily. On the LP side, state-owned platforms accounted for 41% of H1 commitments by Zero2IPO's count.
This paradigm holds under one condition only: acquired or in-house service businesses see margins lift because of AI. If not, it is leveraged consolidation dressed as venture capital, capped by the cost of capital rather than capacity. Public readings in two columns.
Supports
Against
This page is updated on an event basis, refreshed after each platform's funding, acquisition or regulatory disclosure. US figures come from company announcements and reporting by TechCrunch, PitchBook, CNBC and Newcomer; funding amounts and valuations are shown separately and cannot be added; "acquired" counts are self-reported by each platform. China AI funding uses the ITjuzi (AI-specific) basis, and state-capital share uses the Zero2IPO (all VC/PE) basis; the two cannot be added to or subtracted from each other. Funding amounts and valuations for leading model companies follow the original reports, and the same round may differ across sources (e.g., DeepSeek CNY 51bn vs. CNY 50bn); reported valuation ranges for StepFun are inconsistent and are not used here. Margin evidence marked "per GC" is General Catalyst's own claim, not independently verified. This page is a research tool and personal notes, not investment advice.