Overseas Report Positions Zhipu as "China's Palantir," MiniMax as Comparable to Character AI, and Doubao as Analogous to ChatGPT
N.R. Finch
An anonymous report reshared by Marc Andreessen maps Zhipu, MiniMax, and Doubao onto Palantir, Character AI, and ChatGPT — three diverging paths that are rewriting how China's AI companies get valued.
Why does this report matter?
Prominent VC Marc Andreessen reshared a report by anonymous researcher FleetingBits, mapping China's top AI players beyond DeepSeek.
The central thesis: China's large-model race has moved past technology convergence and into full-blown business-model divergence.
This means → the market is no longer just asking "whose model is better" — it is asking who can turn a model into revenue.
Why is Zhipu AI called "China's Palantir"?
Zhipu's core model: win state-enterprise contracts, then deploy its GLM models (including the latest GLM 5.2) on the client's own servers for deep government and enterprise services.
In plain terms = Palantir makes money by building data systems for the U.S. government and large corporations. Zhipu runs the same playbook — installing AI "inside the client's own walls."
The financials back it up: full-year gross margin hit 41%, revenue grew 131.9% year-on-year to RMB 724 million, the highest among China's large-model companies.
Zhipu's losses widened — why is the market still buying in?
The company raised API prices three times within the year and announced a further 83% hike for Q1 2026; open-platform API annualized recurring revenue (ARR) has reached roughly $250 million.
Despite a 60% year-on-year increase in net losses for 2025, Zhipu's Hong Kong market cap broke HK$1 trillion, with shares up more than 1,900% for the year.
This means → investors are placing pricing power and volume-price momentum above near-term profitability. The ability to keep raising prices without losing clients matters more than whether the company is profitable today.
Where does MiniMax's revenue actually come from?
The report positions MiniMax as a Character AI counterpart, focused on consumer companionship and content generation through AI companion app Talky and video-generation app Hailuo.
The key data point: most of MiniMax's revenue comes from outside China, with a significant share from the U.S. market.
In plain terms = while domestic players fight a brutal traffic war, MiniMax chose to earn abroad — where users' willingness to pay is higher.
Despite a full-year net loss of $1.9 billion, shares rose 13% after results, reflecting investor buy-in on the "AI apps going overseas" thesis.
Why is Doubao called "China's ChatGPT"?
The report states explicitly: ByteDance's Doubao is currently China's most popular AI application, effectively playing the role of "China's ChatGPT."
Unlike some competitors, ByteDance maintains a firm closed-source strategy in China — it does not open-source any of its models.
This means → closed source lets ByteDance keep core technology and data tightly in-house, build a walled ecosystem, and maximize Doubao's consumer monetization potential.
The report also notes that ByteDance has not made a big splash in the U.S. market precisely because it does not open-source — a deliberate trade-off, not a capability gap.
What does this three-way divergence mean?
Zhipu = government and enterprise on-premise deployment, building a moat on gross margins and pricing power.
MiniMax = consumer apps going overseas, building a moat on offshore revenue scale.
Doubao = closed ecosystem, building a moat on a walled garden and traffic monetization.
In plain terms = China's AI commercialization race has entered its second half. The question is no longer "whose model is smarter" but "who builds a quantifiable moat first" — and the answer will set the direction of the next revaluation cycle.
Content is for reference only, not financial advice.