China’s AI wave pivots: founders abandon “who’s OpenAI” race for verticals and Palantir-style plays

From OpenAI mimicry to pragmatic pivots

The leadership of China’s AI “six tigers” are quietly changing the script. Li Kaifu (李开复), founder of ZeroOne Wanwu (零一万物), and Wang Xiaochuan (王小川), CEO of Baichuan Intelligence (百川智能), have both signalled a move away from the feverish pursuit of general-purpose foundation models—what many once called “the China OpenAI” moment—toward revenue, specialization and survivable business models. Why the shift? Because the math of winning is no longer just clever models; it’s GPUs, power, data, and capital—and those resources are concentrated with a few U.S. giants.

Li Kaifu’s internal message, leaked after ZeroOne Wanwu’s third‑anniversary note, reportedly reframed the company to benchmark against Palantir rather than OpenAI, with new targets of profitability in 2026 and an IPO in 2027; it has been reported that ZeroOne’s cumulative orders already exceed RMB 1.5 billion. Wang Xiaochuan has been even more blunt in interviews, warning that the pretraining dividend is fading and that China’s foundational models are widening their gap with U.S. counterparts. Baichuan, by contrast, has gone “all in” on healthcare—launching a new medical model and an AI family doctor product—and reportedly holds nearly RMB 3 billion in cash, making its bet more strategic than desperate.

The squeeze from U.S. incumbents and China’s comparative advantage

This recalibration comes against a backdrop of massive capital expenditure from U.S. cloud and AI backers—Microsoft, Google, Amazon and Meta are planning capital outlays measured in hundreds of billions of dollars—which resets the industry every few months with each new OpenAI or Sora upgrade. It has been reported that these four companies will spend more than US$725 billion combined on capex in the relevant window. China’s internet giants are racing to catch up—ByteDance recently raised its 2026 AI capex target from RMB 160 billion to RMB 200 billion, Alibaba has pledged multibillion‑dollar cloud and AI investments, and Tencent’s quarterly capex has already jumped into the hundreds of millions of yuan—but the gap in GPU supply chains and foundational research remains a strategic constraint, especially amid export controls and broader U.S.–China tech competition.

The market consequences are visible. Zhipu AI (智谱) and MiniMax rode a period of tech and policy tailwinds to public listings and sky‑high valuations, but those windows appear to be closing. Meanwhile alternative plays—Moon’s Dark Side (月之暗面), Jieyue (阶跃), DeepSeek and others—are carving distinct paths: some double down on open source and base‑level innovation, others chase verticals or B2B products that can monetize without trillion‑dollar capex.

What this means for China’s AI story

The deeper implication is a maturation of China’s AI ecosystem: not every winner will be a foundation model champion. China’s comparative strengths remain scenario depth, application engineering, manufacturing and rapid commercialization at scale. The industry is learning that being “more like OpenAI” is neither necessary nor sufficient to survive. New leaders are betting that narrow, high‑value verticals—healthcare, government and enterprise analytics—or Palantir‑style data services will deliver returns that the general‑model arms race cannot.

So who wins? The short answer: pragmatic companies that convert Chinese strengths into paying customers rather than burning to replicate the U.S. industrial stack. The long answer depends on geopolitics, chip supply and whether Beijing and domestic capital can sustain second‑order investments in AI infra. For now, the era of “who’s most like OpenAI” is giving way to a more sober competition about product, margins and cash.