China's AI Unicorns Are Splitting Apart: The Great Divergence of 2026

*In 2026, China's six AI unicorns took six different paths. IPOs, pivots, massive funding rounds, and quiet disappearances. The divergence says more about China's AI future than any single model release.*

The Fork in the Road

In January 2026, two Chinese AI companies rang the opening bell at the Hong Kong Stock Exchange within days of each other. Zhipu AI listed first, raising $536 million at a valuation just under $7 billion. MiniMax followed a week later, raising $619 million at $6.5 billion. Both were "Six Little Dragons"—the six independent Chinese AI startups that investors tracked as China's best hope for building foundational models to rival OpenAI.

By September 2026, the picture looks nothing like it did in January.

Zhipu AI, now trading as Knowledge Atlas, has seen its market capitalization climb to approximately $93 billion—a roughly 1,300% gain from its IPO price. MiniMax trades at approximately $33 billion, a fivefold increase. Moonshot AI, which stayed private, raised approximately $3.5 billion in 2026, pushing its valuation from roughly $4.3 billion to $35–50 billion, and has reportedly filed confidentially for a Hong Kong IPO. DeepSeek, historically self-funded through its quant-trading parent High-Flyer, is reportedly in talks for its first external funding round at $45–71 billion, led by China's National IC Industry Investment Fund.

And then there are the other two.

Baichuan AI has effectively dissolved as a general-purpose AI company. Founder Wang Xiaochuan pivoted the company entirely to healthcare AI, slashed the workforce by more than half, and restructured around medical applications. 01.AI, founded by legendary computer scientist Kai-Fu Lee, abandoned the consumer chatbot market and repositioned as an enterprise AI solutions provider, reporting 1.5 billion yuan in contracted orders and near-profitability.

Six companies. Six radically different trajectories. One industry that is no longer moving in a single direction.

The Six Dragons: Who They Were

The "Six Little Dragons" (六小虎) label emerged in Chinese tech media around 2023 to describe the six most prominent independent AI startups building large language models from scratch: Zhipu AI, Moonshot AI, MiniMax, Baichuan AI, 01.AI, and StepFun. (Some versions include DeepSeek as a seventh member; others treat DeepSeek separately because of its unique quant-fund origins.)

These six companies shared a common origin story. All were founded between 2019 and 2023 by researchers or entrepreneurs with elite pedigrees. All raised billions from the same investor pool: Alibaba, Tencent, HongShan, and government guidance funds. All pursued the same strategy: build a large foundation model, release it, and scale toward revenue.

For two years, the narrative was unified. China had its "Six Dragons" racing against each other and against OpenAI. That narrative collapsed in 2026.

Path One: The Public Markets (Zhipu AI and MiniMax)

Zhipu AI and MiniMax chose the IPO route, listing on the Hong Kong Stock Exchange in January 2026. Their decisions were driven by a simple reality: private markets for Chinese AI were cooling, and Hong Kong's public markets were hungry for AI stories.

Zhipu AI: From $7 Billion to $93 Billion

Zhipu AI raised $536 million at roughly $6.8 billion, pricing shares at HK$43.15. By September 2026, the stock had risen approximately 1,300%, giving it a market cap of roughly $93 billion.

What drove this? Zhipu AI's GLM-5 model received strong benchmark scores on coding and reasoning. The company's enterprise API business reportedly reached $150 million in annualized revenue by Q2 2026. And the scarcity of Hong Kong-listed Chinese AI stocks created a feedback loop of demand.

The risk is that Zhipu AI's stock has far outrun its fundamentals. At $93 billion, the bet investors are making is that Zhipu AI will become one of China's two or three surviving foundation model providers.

MiniMax: The Talkie Effect

MiniMax's IPO was driven less by model benchmarks and more by Talkie, an AI companion app that reached 50 million monthly active users globally by mid-2026—a rare case of a Chinese AI product achieving significant traction in Western markets.

MiniMax raised $619 million at $6.5 billion. By September 2026, its market cap had risen to approximately $33 billion.

MiniMax's challenge is proving that Talkie's consumer success is durable. The AI companion market is crowded, and Talkie's global reach makes it vulnerable to regulatory changes in Western markets.

Path Two: The Private Giants (DeepSeek and Moonshot AI)

While Zhipu and MiniMax went public, DeepSeek and Moonshot AI chose to stay private—and grew even larger.

DeepSeek: The $71 Billion Question

DeepSeek is the most unusual company in China's AI landscape. Founded in Hangzhou in 2023 as a spinout from High-Flyer Quant, DeepSeek was entirely self-funded for its first three years. The DeepSeek-R1 release in January 2025 was a global inflection point: the model matched GPT-4 on reasoning benchmarks while costing an estimated $5.6 million to train—less than 1/100th of GPT-4's reported cost. The release triggered an 18% drop in Nvidia's stock price.

For 18 months after R1, DeepSeek stayed private and self-funded. But in mid-2026, reports emerged that DeepSeek was in talks for its first external round, targeting $3–4 billion at $45–71 billion, led by China's National IC Industry Investment Fund.

The significance goes beyond the numbers. For three years, DeepSeek was proof that Chinese AI could compete without Silicon Valley-style venture capital. The decision to raise from a state-linked fund suggests the cost of staying at the frontier has grown beyond what even a successful quant fund can sustain.

Moonshot AI: From $4.3 Billion to $50 Billion in Nine Months

Moonshot AI's 2026 is the most dramatic funding story in global AI. The company began the year valued at roughly $4.3 billion. By September, it had reportedly closed approximately $3.5 billion in new funding, pushing its valuation to $35 billion and reportedly seeking a new round at a $50 billion pre-money valuation.

The catalyst was Kimi K3, released in July 2026. The model featured a 2.5 trillion parameter architecture with a 2 million token context window—the longest of any commercially available model. Benchmark scores were competitive with GPT-5 and Claude Opus 4.7, and the model's agent-swarm capabilities attracted significant enterprise interest.

Moonshot AI has also reportedly filed confidentially for a Hong Kong IPO. At $35–50 billion, the company is priced for perfection. Any slip in model quality or revenue growth could trigger a painful repricing.

Path Three: The Pivots (Baichuan AI and 01.AI)

Not every dragon flew. Two of the six took a different path entirely—abandoning the foundation model race and reinventing themselves as vertical specialists.

Baichuan AI: The Healthcare Pivot

Baichuan AI's story is the most dramatic reversal. Founded in 2023 by Wang Xiaochuan, former CEO of Sogou, Baichuan AI raised over $300 million in its first year and was valued at over $1 billion. But by early 2026, the company was struggling to differentiate against DeepSeek, Moonshot, Zhipu, and the big tech labs. In a decisive move, Wang pivoted the company entirely to healthcare AI, slashing the workforce by more than 50%.

The pivot was described as a "hard landing"—rapid and painful, with hundreds laid off and long-term projects canceled. But Wang was pragmatic. "The general model race is over," he reportedly told staff. "We either find a niche where we can win, or we die."

China's healthcare AI market is large and underserved. Baichuan AI has reportedly signed contracts with several provincial hospitals. But the company is no longer a contender for China's AI crown.

01.AI: The Enterprise Pivot

01.AI's pivot was less dramatic than Baichuan's but equally significant. Founded by Kai-Fu Lee, 01.AI initially pursued the same strategy as the other dragons: build an open-weight foundation model (the Yi series), release it to developers, and scale toward revenue.

But by late 2025, Lee recognized that 01.AI could not compete with DeepSeek on cost, Moonshot on context length, or Zhipu on enterprise traction. He repositioned 01.AI as an enterprise AI solutions provider—effectively, "China's Palantir"—offering customized AI systems for specific industries.

The repositioning has worked. 01.AI reportedly has 1.5 billion yuan in contracted orders and is approaching profitability—a rarity among Chinese AI startups. Lee's decision reflects a broader trend: the separation of "model builders" from "model users." The model builders are in a capital-intensive arms race requiring billions. The model users can build valuable businesses by applying existing models to specific problems.

StepFun: The Sixth Dragon's Uncertain Path

StepFun, the sixth dragon, sits in an ambiguous position. The company has raised approximately $2.5 billion in 2026 and is reportedly restructuring for a Hong Kong IPO. But it has not achieved the breakout product success of MiniMax's Talkie, the technical recognition of DeepSeek's R1, or the enterprise traction of Moonshot's Kimi.

StepFun's situation illustrates the narrowing window for Chinese AI startups. In 2023–2024, there was room for six dragons. In 2026, the market is consolidating around three or four leaders.

The Capital Map: Where the Money Flowed in 2026

The divergence of China's AI unicorns is reflected in their funding patterns. 2026 was the year Chinese AI startups raised more capital than ever before—but that capital was concentrated in an increasingly small number of companies.

A critical shift in 2026 was the changing composition of investors. Three years ago, Chinese AI startups were funded primarily by venture capital firms. In 2026, the dominant capital sources are:

1. Big Tech Strategics: Alibaba, Tencent, and Meituan have become the most active AI investors.

2. State-Linked Funds: The National IC Industry Investment Fund ("Big Fund") and provincial guidance funds now account for an estimated 40% of AI VC activity in China—up from 20% three years ago.

3. International Investors: Largely absent. US export controls and geopolitical tensions have made most Western VCs cautious.

The shift toward state-linked funding is particularly significant. For the first time, China's government is directly underwriting frontier AI model development through vehicles like the National IC Industry Investment Fund. This represents a strategic decision to treat AI as a national priority on par with semiconductors.

What the Divergence Means for Global AI

The splitting of China's AI unicorns is not just a business story. It has implications for the global competition over artificial intelligence.

First, China's AI ecosystem is maturing—and consolidating. In 2023, there were six credible independent AI startups. In 2026, there are effectively three: DeepSeek, Moonshot, and Zhipu (with MiniMax as a consumer-focused fourth). The foundation model market is following the same pattern as search engines and social media: fragmentation followed by rapid consolidation around a few winners.

Second, Hong Kong has become the liquidity hub for Chinese AI. With US listings effectively closed to Chinese AI companies since 2023, Hong Kong has emerged as the primary exit route. Zhipu and MiniMax's successful debuts have established a template, and Moonshot's anticipated IPO will test whether the market can absorb another mega-listing.

Third, the "open weights" strategy is becoming China's deliberate export. DeepSeek's MIT-licensed models, Zhipu's open releases, and Moonshot's developer-friendly APIs are all part of a strategy to establish Chinese AI as the default choice for developers in regions that cannot afford Western API prices.

Fourth, the pivoters—Baichuan and 01.AI—may be the canaries in the coal mine. Their decisions to abandon the foundation model race suggest that even well-funded startups are finding the economics unsustainable. If training costs continue to rise while revenue lags, more companies may follow their lead.

The Unanswered Questions

For all the clarity that 2026 has brought, significant uncertainties remain.

Can the public market valuations be sustained? Zhipu AI at $93 billion and MiniMax at $33 billion are pricing in years of exponential growth. If revenue growth slows—or if a new model from DeepSeek or Moonshot leapfrogs their offerings—these stocks could retrace sharply.

Will DeepSeek's first external round change its culture? DeepSeek's identity has been built on independence—freedom from investor pressure, freedom to publish, freedom to optimize for research rather than revenue. State-linked investors may bring different expectations.

What happens to the pivoters? Baichuan's healthcare AI and 01.AI's enterprise solutions are sensible strategic retreats. But are they large enough to support companies that once aspired to build China's answer to GPT? Or will they be acquired, merge, or simply fade?

Will Hong Kong remain open? The success of Zhipu and MiniMax's IPOs depends on continued investor appetite for Chinese AI. A geopolitical shock or market downturn could close the IPO window as quickly as it opened.

Can China build AI without the best chips? US export controls continue to block access to H100/H200-class GPUs. Chinese labs have compensated with software optimization, but the gap between available hardware and frontier requirements is widening.

Social Media Reactions

*The following comments were collected from Chinese social media and tech forums in September 2026. They are presented in the original Chinese with English translations.*

@投行分析师老张 (Weibo)

"智谱从7亿到93亿市值，这不是估值，这是信仰。问题是，信仰能撑多久？"

*("Zhipu went from $700M to $93B market cap. That's not valuation, that's faith. The question is: how long can faith last?"*)

@AI产品经理小林 (Zhihu)

"六小虎变三龙，百川和零一都转型了。这说明基础模型这个赛道，真的只能活三家。"

*("Six dragons became three. Baichuan and 01.AI both pivoted. It shows the foundation model race can only support three survivors."*)

@硅谷观察者 (Twitter/X)

"DeepSeek raising at $71B from China's Big Fund is the moment when Chinese AI stopped being a startup story and became a state project."

*(No translation needed—comment was posted in English.)*

@港股小散户 (Xueqiu)

"买了智谱和MiniMax，今年收益跑赢99%的基金经理。但每天睡觉都不踏实，怕哪天一个监管文件就跌停。"

*("I bought Zhipu and MiniMax. This year's returns beat 99% of fund managers. But I can't sleep peacefully—afraid one regulatory document could trigger a limit-down."*)

@AI研究员阿明 (Zhihu)

"月之暗面从4亿到50亿估值只用了9个月，这比当年拼多多还快。Kimi K3确实强，但这个速度正常吗？"

*("Moonshot went from $400M to $5B valuation in 9 months—that's faster than Pinduoduo. Kimi K3 is strong, but is this speed normal?"*)

@医疗AI从业者 (WeChat)

"王小川把百川全部转医疗，很多人骂他没骨气。但我觉得这是最理性的选择。通用大模型烧不起，医疗至少有付费方。"

*("Wang Xiaochuan pivoting Baichuan entirely to healthcare—many call him cowardly. But I think it's the most rational choice. General models are too expensive to burn; healthcare at least has paying customers."*)

Sources and Methodology

This article synthesizes reporting from Bloomberg, the South China Morning Post, ChinaBizInsider, StartupHub.ai, NeuronFeed, SecondTalent, ExplainX.ai, and InsideAI. Valuation and funding data are drawn from press reports and industry databases; precise figures for private companies are approximate and subject to revision. Market capitalization data for public companies reflects September 2026 trading prices.

Company-specific data on headcount, revenue, and product metrics is sourced from press reports, founder statements, and industry analysis. The "Six Little Dragons" terminology is drawn from Chinese tech media usage (六小虎); the "Four Dragons" label (四小龙) is a more recent investor categorization referring to DeepSeek, Zhipu, MiniMax, and Moonshot.

*Last updated: September 10, 2026*