Key Takeaways:
Three Chinese AI companies are proving that the path to profitability in artificial intelligence runs through government contracts, overseas consumers, and closed ecosystems — not just better models.
Zhipu AI's government-focused deployment model delivers roughly 40% gross margins and a 1,900% stock surge this year, while MiniMax generates most of its revenue from overseas consumers and ByteDance's Doubao dominates China's consumer market through a closed-source strategy that keeps its technology proprietary.
"The market is finally differentiating between AI companies that can monetize and those that can't," said FleetingBits, an anonymous researcher whose report was shared by venture capitalist Marc Andreessen. "Zhipu has proven that government clients will pay a premium for on-premise deployment, while MiniMax shows that Chinese AI apps can win abroad."
Zhipu AI reported 2025 revenue of 724 million yuan ($99 million), up 131.9% from a year earlier, with gross margin at 41%. Its API platform reached about $250 million in annualized recurring revenue, supporting three price increases in the past year — including an 83% jump in first-quarter 2026 pricing. The company's Hong Kong-listed shares have surged more than 1,900% year to date, pushing its market capitalization past 1 trillion Hong Kong dollars ($128 billion).
The divergence shows that China's AI sector has moved beyond the uniform losses that defined 2024. Investors are now paying premiums for companies with clear monetization paths, even when near-term losses persist. Zhipu's net loss widened 60% in 2025, yet its stock continued climbing — a pattern that mirrors how markets treated early-stage US AI companies before they turned profitable.
Zhipu AI's business model most closely resembles Palantir Technologies, the US data analytics firm known for government contracts. The company deploys its GLM series of models — including the recently released GLM 5.2 — on client-owned hardware, giving state-owned enterprises and government agencies full control over their data.
This approach has produced margins that few AI companies anywhere can match. Zhipu's 41% gross margin compares with roughly 50% to 60% at OpenAI and Anthropic, which operate cloud-based services with higher infrastructure costs. The trade-off is scale: Zhipu's $99 million in annual revenue is a fraction of OpenAI's estimated $3.7 billion in 2024 revenue.
GLM 5.2, released last week, has drawn attention for its performance on agentic benchmarks. The open-source model scores within 1 percentage point of Anthropic's Opus 4.8 on one closely watched agentic benchmark while costing roughly one-fifth as much to run, according to OpenRouter data. Developer traffic on the platform has climbed faster than it did after DeepSeek's V4 launch in April.
The open-source nature of GLM 5.2 adds another dimension to Zhipu's strategy. Enterprises can download, fine-tune, and run the model on their own servers, making it immune to the kind of government revocation that forced Anthropic to pull its Fable Mythos-class model. OpenAI said Friday it is limiting its GPT 5.6 models because of a government request.
MiniMax has chosen a path that avoids China's crowded domestic market entirely. The company's AI companion app, Talky, and video generation tool, Hailuo, generate most of their revenue from users outside China, including a significant portion from the United States.
The strategy carries risks. US regulators have shown increasing willingness to restrict Chinese technology products, and MiniMax's $1.9 billion net loss in 2025 raises questions about how long it can sustain overseas expansion without additional funding. Still, the company's stock rose 13% after its earnings release, suggesting investors see the overseas revenue stream as a valuable hedge against domestic competition.
ByteDance's Doubao, described by FleetingBits as "China's ChatGPT," takes the opposite approach. The app is the most popular AI application in China, but ByteDance has chosen not to open-source any of its models — a decision that limits its global profile but protects its technology and data.
"ByteDance doesn't open source anything, which is why it hasn't caused the same stir in the US as DeepSeek," FleetingBits wrote. The closed-source strategy allows ByteDance to maintain tight control over its product experience and monetization, maximizing Doubao's revenue potential in China's 1.1 billion-user internet market.
The three models carry different risk profiles. Zhipu's government focus provides stable, high-margin revenue but limits total addressable market to China's public sector. MiniMax's overseas bet offers scale but exposes the company to regulatory and geopolitical headwinds. ByteDance's closed ecosystem maximizes domestic revenue but cedes the global open-source race to competitors like Alibaba and DeepSeek.
Zhipu AI trades at roughly 80 times trailing revenue, a multiple that reflects investor enthusiasm for its government contract model rather than current earnings. The company's ability to sustain its 41% margin as more Chinese AI startups pursue similar government deals will determine whether that valuation holds.
This article is for informational purposes only and does not constitute investment advice.
