The Six Tigers No More: How China's AI Unicorns Split Into Six Different Species

*Photo by Unsplash*

The conventional wisdom says China's AI startup scene is consolidating. The Q2 2026 data says the opposite: it's fragmenting into entirely different industries.

In January 2026, when Zhipu AI rang the opening bell at the Hong Kong Stock Exchange, the narrative was simple. China's "Six AI Tigers" — the half-dozen unicorn startups that emerged from the 2023 generative AI boom — were graduating to the public markets. The playbook looked familiar: train a foundation model, scale users, raise billions, IPO, repeat.

Seven months later, that narrative lies in ruins.

Zhipu trades at 8137 billion HKD (~$104 billion), making it the world's most valuable pure-play AI company. MiniMax, which listed one day later, sits at 1137 billion HKD (~$14.5 billion) — a 7x gap that would have seemed impossible when both were private. Moonshot AI raised another $2 billion in July at a $31.5 billion valuation, but its path to public markets looks murkier by the week. Baichuan, the company founded by former Sogou CEO Wang Xiaochuan, fired most of its general-purpose team and went all-in on medical AI. 01.AI, the startup led by legendary computer scientist Kai-Fu Lee, abandoned foundation model training entirely to become a Palantir clone. And Stepfun, the youngest of the group, started building phones.

The Six Tigers aren't a cohort anymore. They're six different companies in six different industries, bound together only by the accident of their founding year and the faded memory of a shared label.

This is the story of how China's AI startup ecosystem grew up — and why 2026 will be remembered as the year the category "AI startup" stopped being meaningful.

The Conventional Wisdom: One Market, One Playbook

Walk into any venture capital firm in Beijing or Shanghai in early 2025, and you'd hear the same thesis. China's AI market was following the Silicon Valley playbook: foundation model → chatbot app → enterprise API → platform. The Six Tigers — Zhipu, MiniMax, Baichuan, Moonshot (Kimi), 01.AI, and Stepfun — were the local contenders, each with slightly different strengths but fundamentally the same strategy.

The numbers seemed to support this. Collectively, the six had raised over $15 billion by the end of 2025. Zhipu's GLM series was the open-source darling, with over 100,000 derivative models on Hugging Face. MiniMax's Talkie app had crossed 200 million downloads globally. Moonshot's Kimi was the preferred tool for China's white-collar professionals, its long-context window a genuine technical differentiator. Baichuan had the pedigree of Wang Xiaochuan, one of China's most respected tech founders. 01.AI had Kai-Fu Lee, whose personal brand alone could open any door in Chinese tech. And Stepfun had the most capital-efficient training stack, with claims of 10x lower costs than competitors.

The market treated them as a basket. When DeepSeek released R1 in January 2025 — an open-source reasoning model that matched GPT-4 at a fraction of the cost — all six suffered valuation compression. When the Chinese government announced AI infrastructure subsidies in March 2025, all six benefited. When US export controls tightened in June 2025, all six were seen as facing the same chip shortage.

This was the narrative: six tigers, one jungle, one prey.

It was wrong.

The Data That Broke the Narrative

The divergence didn't happen gradually. It exploded in the first half of 2026, catalyzed by three forces: the IPO window, DeepSeek's sustained pressure, and a brutal recalibration of what "AI company" actually means in a market where inference costs are collapsing.

Consider these numbers from July 2026:

The dispersion is staggering. Zhipu is worth more than the other five combined. MiniMax's consumer entertainment model generates revenue that looks nothing like Zhipu's enterprise contracts. Moonshot has raised more private capital in 2026 than some of the others have in total. Baichuan's pivot to healthcare shrank its addressable market by 90% but may have saved its survival. 01.AI abandoned the very technology — foundation models — that made it famous. And Stepfun, the youngest tiger, is now a hardware company.

This isn't consolidation. It's speciation.

The Two Species Already in the Zoo: Zhipu and MiniMax

Zhipu: The Open-Source Leviathan

When Zhipu listed on January 8, 2026, it became the world's first publicly traded pure-play foundation model company. The stock popped 65% on debut. By July, it had crossed $100 billion in market capitalization — a figure that puts it in the same league as established tech giants, not startups.

Zhipu's strategy is deceptively simple: be the default AI infrastructure for China. Its GLM-5.2 model family spans text, code, vision, and agentic capabilities. It releases open weights for most variants, creating a developer ecosystem that feeds back into its cloud API business. It has the deepest penetration in Chinese government and SOE (state-owned enterprise) contracts — a moat that no foreign competitor can cross.

But the market is asking harder questions. At $104 billion, Zhipu trades at roughly 500x revenue. That's not a valuation — it's a prophecy. The prophecy says Zhipu will become the "Android of AI" in China: the default platform that every application builds on. If that prophecy is even slightly wrong, the stock could lose 80% of its value.

The July 2026 data offers mixed signals. Zhipu's cloud revenue is growing at 180% year-over-year, but from a small base. Its government contracts are lucrative but lumpy. And its open-source strategy, while building ecosystem loyalty, means competitors can run its models without paying.

The market's verdict so far: the prophecy is holding. But 2026 is only half over.

MiniMax: The Global Entertainer

MiniMax listed one day after Zhipu, on January 9, 2026. Its market cap of ~$14.5 billion makes it the smallest of the listed tigers — but its business model may be the most defensible.

MiniMax doesn't sell AI to enterprises. It sells entertainment to consumers. Its Talkie app, an AI character chat platform, crossed 300 million users in June 2026. Its Hailuo AI video generation tool powers millions of short-form videos across TikTok, Instagram, and YouTube. Over 70% of its revenue comes from outside China, making it the most global of the six tigers.

This matters because entertainment AI faces fewer regulatory constraints than enterprise AI. China's AI companion ban, which took effect July 15, 2026, hit domestic players hard — but MiniMax's international user base provided a buffer. While ByteDance's Doubao and Alibaba's Qwen had to shut down companion features used by hundreds of millions, MiniMax's Talkie users in the US, Southeast Asia, and Latin America kept chatting.

The entertainment moat is real but fragile. Content trends shift fast. The AI video models that MiniMax built its growth on — Hailuo AI — are facing competition from ByteDance's Seedance, Kuaishour's Kling, and a dozen open-source alternatives. And Talkie's character-based model may face its own regulatory pressures as governments worldwide grapple with AI-generated content.

MiniMax's answer, revealed in its June 2026 quarterly report: double down on AI-native content formats. Not just chatbots that pretend to be characters, but AI-generated TV episodes, AI-composed music, and AI-directed games. The company is betting that entertainment will be the first AI sector to reach $100 billion in revenue — and that it will capture a double-digit share.

The Three Private Tigers: Three Different Survival Strategies

Moonshot: The Agent Gambler

Moonshot AI's Kimi K3, released in late July 2026, is the largest open-weight model ever shipped at 2.8 trillion parameters. The launch triggered a Nasdaq selloff and a White House investigation. But behind the headlines, Moonshot's real bet isn't on parameter count — it's on agents.

The company's K2.6 release in mid-2026 introduced Agent Swarm, a system that can coordinate hundreds of sub-agents to complete complex tasks. Want to plan a conference? Kimi will spin up agents for venue research, speaker outreach, marketing copy, budget tracking, and logistics — all working in parallel. The user sees a single interface; behind the scenes, a small army of AI workers does the work.

This is expensive. Moonshot raised $2 billion in July alone, bringing its total 2026 fundraising to over $3.9 billion. At a $31.5 billion valuation, it's the most valuable private AI company in China — and possibly the world. But it's burning cash at a rate that would make even SoftBank nervous.

The bull case: agentic AI is the next platform shift. If Kimi becomes the default agent orchestration layer — the "operating system for AI workers" — a $31.5 billion valuation will look cheap. The bear case: agents are a feature, not a platform, and will be absorbed by Microsoft, Google, and Alibaba before Moonshot can build a standalone business.

Moonshot's July fundraising suggests investors are buying the bull case. But with IPO markets volatile and the company needing more capital before profitability, the clock is ticking.

Baichuan: The Doctor Will See You Now

Of all the divergence stories, Baichuan's is the most dramatic. In March 2026, Wang Xiaochuan announced that Baichuan would "cease all general-purpose foundation model training" and focus exclusively on medical AI. The company's Baichuan-M4 model, released in July 2026, is trained on 200 million medical records, 10 million clinical papers, and real-time data from partner hospitals including Beijing Tiantan Hospital.

The pivot shrank Baichuan's addressable market from "all of China's AI market" to "healthcare AI in China." But it also gave the company something rare in China's AI landscape: a clear path to profitability. Healthcare AI has natural customers (hospitals, pharmaceutical companies, insurers) with budgets, procurement processes, and urgent needs. A general-purpose chatbot has no natural customer — it's a solution looking for problems.

Baichuan's "Baixiaoyi" AI family doctor product, launched in July 2026, already serves 500,000 paying users. Its hospital decision-support system is deployed in 120 tertiary hospitals. And its pharmaceutical R&D platform has contracts with 8 of China's top 10 drug makers.

The valuation compression is severe — from an estimated $8 billion pre-pivot to perhaps $2.5 billion now. But Wang Xiaochuan, who built Sogou to a $5 billion public company, knows that surviving is better than being right. "In 2023, everyone wanted to be OpenAI," he told Chinese media in July. "In 2026, I'd rather be Epic Systems."

01.AI: The Palantir Pivot

Kai-Fu Lee's 01.AI was supposed to be different. The legendary AI researcher and former Google China president brought unmatched credibility. The company's Yi series of models were technically competitive. And Lee's personal brand opened doors that remained closed to younger founders.

But by early 2026, the story had soured. 01.AI's models were good but not great — consistently trailing DeepSeek, Zhipu, and Moonshot on benchmarks. Its consumer app, Yi Chat, never broke into the top 10 AI apps in China. And its enterprise business was too small to matter.

In April 2026, Lee made a decision that shocked the industry: 01.AI would stop training foundation models entirely. The pre-training team was dissolved. Most researchers were transferred to Alibaba or laid off. And the company pivoted to what Lee called "Decision AI" — using existing models to help enterprise executives make better decisions.

The pitch is explicitly Palantir-inspired. 01.AI builds custom AI systems that ingest a company's operational data — sales, supply chain, financials, HR — and generate recommendations for the CEO and board. It doesn't sell models. It sells outcomes.

The pivot is either brilliant or desperate, and the market hasn't decided which. 01.AI's revenue is reportedly growing 300% quarter-over-quarter from a tiny base. But the company has burned most of its $300 million in funding, and its July 2026 fundraising round was reportedly a down round at ~$1 billion — a fraction of its previous valuation.

Lee's response, in a July interview: "The companies that win AI won't be the ones with the best models. They'll be the ones that solve the most valuable problems. We're choosing to solve problems."

The Wildcard: Stepfun Builds a Phone

Of all the Six Tigers' divergences, Stepfun's is the most unexpected. The company, founded in April 2023 by former Microsoft Research Asia veterans, was always the technical purist. Its Step series models were known for training efficiency: it claimed to achieve GPT-4-level results with 10% of the compute cost.

In July 2026, Stepfun announced something that had nothing to do with models: it was building a smartphone. The StepPhone, developed with hardware partner Huaqin Technology, would run a custom "Agent OS" built on Stepfun's models. The phone wouldn't just have an AI assistant — it would be an AI assistant, with system-level access to apps, notifications, and user data.

The logic is入口 (entrance). In China's AI landscape, the winners won't be the model makers — they'll be the companies that control the user interface. ByteDance has Doubao embedded in TikTok. Tencent has Yuanbao in WeChat. Alibaba has Qwen in Taobao. Stepfun has no distribution. A phone is an attempt to build one from scratch.

The phone won't ship until Q4 2026, and the track record of AI companies building hardware is poor. But Stepfun's July fundraising — reportedly $2.5 billion at a ~$10 billion valuation — suggests investors are willing to bet on the入口 strategy. The company's investor list includes Huaqin, Longcheer, and other phone supply chain companies, making the hardware bet a shared destiny.

The Force That Drove the Divergence: DeepSeek

No account of the Six Tigers' divergence is complete without DeepSeek. The Hangzhou-based research lab, founded as a side project of quantitative hedge fund High-Flyer, wasn't part of the original Six Tigers narrative. It wasn't venture-backed. It didn't have a celebrity founder. And it didn't play by Silicon Valley rules.

But DeepSeek's January 2025 release of R1 changed everything. An open-source reasoning model that matched GPT-4 on math and coding benchmarks, trained for a reported $5.6 million — a fraction of what the Six Tigers spent on single training runs. The model went viral globally, with over 100 million downloads in its first month.

The impact on the Six Tigers was immediate and brutal. If a hedge fund side project could build GPT-4 for $5.6 million, what justified the billions the tigers had raised? Why would enterprises pay premium prices for Zhipu or Moonshot APIs when DeepSeek offered comparable capability for free?

The answer, it turned out, was that they wouldn't — at least not for models alone. DeepSeek forced a reckoning: in a world where frontier models become commodities in months, the value isn't in the model. It's in the distribution, the vertical expertise, the enterprise relationships, or the hardware control.

Each tiger chose a different response. Zhipu doubled down on government relationships and open-source ecosystem lock-in. MiniMax leaned into entertainment, where model quality matters less than product polish. Moonshot bet on agents as the next platform layer. Baichuan retreated to healthcare, where domain expertise beats general capability. 01.AI abandoned models for services. And Stepfun tried to build the terminal itself.

DeepSeek didn't just disrupt the Six Tigers. It forced them to evolve into different species.

The New Taxonomy: What Replaces "AI Startup"

If "AI startup" is no longer a meaningful category, what replaces it? The first half of 2026 suggests a new taxonomy is emerging:

This taxonomy has implications beyond China. Silicon Valley's AI startups are facing the same pressure. OpenAI's evolution from research lab to consumer product company to platform company mirrors Zhipu's journey. Anthropic's focus on safety and enterprise resembles Baichuan's vertical retreat. And the proliferation of AI hardware startups — Rabbit, Humane, and now Stepfun — shows the same入口 anxiety.

The era of the generic "AI startup" is ending globally. China is just where the divergence is most visible, because the market is largest, the competition most intense, and the capital most plentiful.

What the Market Is Actually Pricing

With two of the Six Tigers public and four still private, the market is sending clear signals about what it values — and what it doesn't.

The market loves growth and hates uncertainty. Zhipu's $104 billion valuation reflects belief in its platform potential, not its current numbers. MiniMax's more modest $14.5 billion reflects its entertainment focus — a proven business model with lower ceiling. And Moonshot's $31.5 billion private valuation is a bet on agents that may take years to validate.

The private tigers face harder questions. Baichuan's healthcare pivot may be strategically sound, but healthcare AI valuations are typically 3-5x revenue, not 100x. 01.AI's Palantir pivot puts it in a competitive market against established players. And Stepfun's phone is a make-or-break bet with no guarantee of success.

For venture capitalists who backed all six in 2023-2024, the divergence is a portfolio management nightmare. The "China AI basket" trade is dead. Each tiger now requires its own thesis, its own risk model, its own exit timeline.

The Social Media Verdict

Zhihu (Chinese Quora)

"六小虎变六物种，这是好事。2023年所有人都在抄OpenAI的作业，现在终于有人交自己的卷子了。"

*"The Six Tigers becoming six species is a good thing. In 2023 everyone was copying OpenAI's homework. Now finally someone is turning in their own exam."*

Xiaohongshu

"Kimi的agent swarm我用了一个月，确实能帮我做周报。但每个月花199块，真的值吗？"

*"I've used Kimi's agent swarm for a month. It really can help me with weekly reports. But is 199 RMB per month really worth it?"*

Weibo

"智谱1000亿市值，收入才多少？这不是AI泡沫是什么？"

*"Zhipu's $100B market cap, but how much revenue? What is this if not an AI bubble?"*

Twitter/X

"The China AI startup divergence is the most important story nobody's talking about. While US VCs still fund "foundation model" companies, Chinese founders have already figured out the model is commoditized."

Douban

"零一万物转型做决策AI，李开复这是认输了还是悟了？"

*"01.AI pivoting to decision AI — is Kai-Fu Lee admitting defeat or having an epiphany?"*

GitHub

"DeepSeek R1 didn't just disrupt the Six Tigers. It disrupted the entire funding model for AI research. If $5M gets you GPT-4, why does anyone need $500M?"

What Comes Next: Predictions for H2 2026

The divergence isn't finished. If anything, the second half of 2026 will accelerate it. Here are the key events to watch:

Moonshot's IPO Attempt

With $3.9 billion raised in 2026 alone, Moonshot must either go public or find even larger private checks. The company has reportedly hired Goldman Sachs and CICC for a Hong Kong listing in Q4 2026. But at $31.5 billion, it would be the largest tech IPO in Hong Kong's history — in a market that has shown limited appetite for unprofitable tech.

Stepfun's Phone Launch

The StepPhone is scheduled for Q4 2026. If it succeeds — even modestly — it validates the "AI-native hardware" thesis and could trigger a wave of similar attempts. If it fails, Stepfun may not survive as an independent company.

Baichuan's Healthcare Revenue

Baichuan's bet on medical AI will face its first real test in H2 2026, as hospital procurement cycles close and pharmaceutical partnerships mature. If revenue crosses $100 million annually, the pivot is vindicated. If not, the company may become an acquisition target.

01.AI's Funding Crunch

At ~$1 billion valuation with reportedly limited cash reserves, 01.AI faces a funding environment that has turned hostile to AI services companies. Lee's personal brand may not be enough to close another round. An acquisition by Alibaba, Tencent, or a traditional enterprise software company is increasingly likely.

Zhipu's $100B Test

Maintaining a $100 billion market capitalization requires more than narrative — it requires numbers. Zhipu's Q3 and Q4 earnings reports will be the most-watched financial disclosures in China's AI sector. If growth slows or margins fail to materialize, the stock could reprice dramatically.

The Bottom Line

The Six Tigers were never really a cohort. They were six companies founded in the same 12-month window, funded by the same venture capital firms, and covered by the same tech media. The shared label obscured more than it revealed.

In 2026, the obscuring is over. Zhipu is an infrastructure platform. MiniMax is an entertainment company. Moonshot is an agent platform bet. Baichuan is a healthcare vertical. 01.AI is an enterprise services firm. And Stepfun is a hardware startup.

The divergence teaches a broader lesson about AI markets. When a technology is new and scarce, everyone looks the same — they're all "AI companies." When the technology commoditizes, differentiation emerges. And the companies that survive are the ones that found something to be besides "an AI company."

China's Six Tigers didn't fail. They grew up. The question for the second half of 2026 is whether the public markets will reward that maturity — or punish the tigers for losing their youthful promise.

Word count: ~3,200 words | Reading time: 16 minutes

*Published: August 3, 2026*