The classic venture capital narrative is "spot the genius, place the bet, and wait." In 2026, that narrative is failing on both sides of the Pacific at once: America's top funds have grown PE hands, buying companies outright, restructuring them, and building new ones from scratch; China's capital, space, and industrial resources are feeding hand-picked contenders into champions. When AI makes the winner-take-all slope steep enough, waiting for a company to grow up before fighting for share makes less sense than stacking the deck before the game even starts.
The most emblematic example is the Thrive family: instead of funding startups, it buys up traditional businesses like accounting firms and IT service providers and rewrites how they work with AI. Raised $2B at a $12B valuation, with OpenAI as a shareholder, and has acquired 70+ companies so far. On Oct 29, it bought NYSE-listed ADT Inc. for $10.00 a share (a 45.6% premium to the market price). The equity is valued at about $7.6 billion, and including assumed debt, the total deal is worth about $15.3 billion — the third such acquisition over $5 billion this year. It's also the first time Thrive has taken an entire public company private, delisting it from the exchange, pending regulatory approval. The most unusual part: OpenAI isn't just a shareholder — it's placing its own staff inside the acquired company. What to watch next: when it buys accounting firms, it keeps the audit license in place while routing profits to the new owner — an arrangement now being tightened on one side and loosened on the other. Whichever rule lands first will decide how costly compliance gets.
General Catalyst takes a more systematic approach: it builds AI companies from scratch through its $1.5B "Creation Strategy", then acquires and integrates legacy players, with a separate $0.75B vehicle handling acquisitions. Public companies already taken: Long Lake (backed by General Catalyst) closed its $6.3 billion acquisition of Amex GBT in late September, delisting it and ending its periodic SEC reporting — whether margins actually improve will depend on the buyer's voluntary disclosures. What to watch: Axiom, spun out of Flex, has closed its $4.4 billion acquisition of EPC Power, but General Catalyst's lead role in a $2 billion preferred-stock financing is still stuck in antitrust review. When retrofitting what already exists is a steadier bet than betting on something new, building your own winner beats picking one.
The winner-making chain is currently stalled, not reversing. Of the 16 chokepoints wangvc tracks, the 7 that have tightened still outweigh the 4 that have loosened, and the other 5 remain undecided. What to watch: Thrive Holdings' Oct 29 take-private of ADT didn't change the state of these chokepoints, but it jumped Thrive's own acquisition scale and risk up by an order of magnitude; DeepSeek's financing round still hasn't been signed and remains stuck in place. Thrive, General Catalyst, and OpenAI are all moves on the same capital network.
Let the data speak first: in H1 2026, China's AI primary-market funding totaled RMB 308B, accounting for 48.6% of all private-market dollars but only 22.5% of deal count — the money is highly concentrated. DeepSeek (¥51B), StepFun (¥23B), and Kimi (¥18.9B) together captured 30%; X Square Robot alone raised ¥6.3B in half a year. The funding window is closing for many model companies. What to watch: DeepSeek's Q4 funding target has been expanded to roughly RMB 100 billion, but the deal still hasn't been signed. Whether that money actually gets signed and closed will decide whether concentration keeps widening this year.
The playbook is even more hands-on: Shenzhen-based Delin Capital spent RMB 40M building an "AI startup utopia" (Delin Residency — that's the facility spend alone), on top of a RMB 2M cash grand prize, up to RMB 5 million in equity investment for each of 15 teams, and 10 weeks of free residency with housing and compute included. The founder's line — "if screening turns up only 3 qualified teams, then there are only 3 slots" — says it all: incubation has moved up to before product validation, and VCs have gone from collecting periodic reports to coaching founders daily. Add in incubation spillover from giants like the SenseTime ecosystem, and China's version of "manufacturing winners" combines money, space, and industrial resources all at once.
AI has made the winner-take-all curve far steeper: model capability doubles with each generation, and early movers' flywheels of data, compute, and talent spin faster and faster. Waiting for a company to grow on its own and then fighting for market share carries far more uncertainty than building the winner yourself. Put simply, building winners means spending money to reduce the uncertainty of "can this be done," rather than betting on "guessing which company will win."
Money is handed out off a list, and companies not on it see their funding windows close fast — that's what more than 200 Chinese model companies are living through right now. The boundaries between VC, PE, and incubator are dissolving too, which means the risk shifts from "picking wrong" to "building wrong": the cost of failing to manufacture a winner is far higher than the cost of missing one.
When OpenAI is simultaneously the model supplier, a shareholder, and the one sending staff into its portfolio companies, the definition of "winning" starts to blur. Is a winner validated by the market, or appointed by capital? The answer to that question determines how much these "manufactured" companies will be worth in the next cycle.
Picking winners pays off on foresight; building winners pays off on execution — the former is a contest of judgment, the latter of whether you can actually pull it off. In the AI era, VCs are shifting from people who pick companies to people who build them.
If General Catalyst's self-built companies don't fetch higher prices at their first sale or IPO than comparable portfolio companies, and the margins of the traditional businesses it acquires don't improve from AI-driven overhauls, that would show this playbook is just the old trick of borrowing money to scale, wearing a "venture capital" label — its ceiling set by the cost of borrowing, not by execution skill.