Key Highlights

  • Zhipu shares plunged over 40% in the past month despite launching GLM 5.2, a top-tier open-weight AI model.
  • Open-weight AI models force labs to bear heavy infrastructure costs while cloud providers capture most inference revenue.
  • Alibaba’s stock rose 13% as its cloud division benefits from hosting third-party AI models.

China’s push to dominate artificial intelligence through open-weight models is collapsing under its own economics. The strategy, championed by President Xi Jinping as a way to challenge U.S. dominance, has instead left Chinese AI startups with soaring infrastructure bills and negligible revenue from their most advanced models.

Shares of Zhipu and MiniMax have plummeted over 40% and 50%, respectively, in the past month, while Alibaba’s stock has climbed 13% as its cloud division profits from hosting third-party AI workloads.

Open-Weight AI Fails the Profit Test

Open-weight AI models, which release trained parameters for public use, differ fundamentally from open-source software. While companies like Red Hat built a $34 billion business by monetizing support and enterprise services around free code, AI labs face a far harsher reality. Every query processed by an open-weight model requires expensive GPUs, electricity, and data-center capacity, turning user growth into a cost burden rather than a revenue driver.

Zhipu, the developer of the acclaimed GLM 5.2 model, lost nearly $500 million last year on just $107 million in revenue, a margin that would bankrupt most startups. MiniMax fared no better, posting a $250 million loss on $79 million in sales. The financial strain is so severe that Moonshot AI, another high-profile lab, had to halt new sign-ups for its Kimi K3 model days after launch due to insufficient computing power.

Unlike software, where marginal costs approach zero, AI models scale at the speed of a lab’s credit line.

Cloud Giants Capture the Value

The economics of open-weight AI tilt overwhelmingly toward cloud providers. These providers charge for compute time while the model creators receive little to no recurring revenue. Analysts at William Blair note that open-weight models generate income primarily through hosting services, but those workloads flow to the most cost-efficient operator, not the model’s creator.

The result is a perverse dynamic where Chinese AI labs subsidize the growth of cloud giants while their own shares crater. Alibaba’s 13% gain over the past month underscores this shift, as its cloud division benefits from hosting third-party models without bearing the upfront training costs. Barclays analyst Raimo Lenshow, who recently toured China’s AI sector, warned that aggressive pricing competition among open-weight models is accelerating commoditization, further squeezing profitability.

Beijing’s Strategic Trade-Off

China’s leadership is prioritizing geopolitical leverage over near-term profits. President Xi Jinping’s recent call for “open-source, openness, collaboration, and sharing” in AI signals a deliberate strategy to pressure U.S. leaders like OpenAI and Anthropic by flooding the market with capable, low-cost alternatives.

Even if Chinese labs like Zhipu and MiniMax never turn a profit, their open-weight models force American competitors to slash prices, making it harder to recoup the billions spent on training. Analysts suggest this approach mirrors China’s broader tech playbook, where market share and technological diffusion take precedence over financial returns. The financial toll is already visible: Zhipu’s 40% share drop and MiniMax’s 50% plunge reflect investor skepticism that open-weight models can ever deliver sustainable margins.

For now, the strategy appears to be working, just not for the companies footing the bill.

The Next Catalyst: Infrastructure or Policy Shift

The immediate pressure point for Chinese AI labs is infrastructure. Moonshot AI’s decision to pause new sign-ups for Kimi K3 highlights the physical limits of scaling open-weight models without massive capital injections. Labs may soon face a choice: either raise billions to build their own data centers or accept permanent dependence on cloud providers.

A more existential risk looms in Beijing’s policy direction. If the government doubles down on openness as a national priority, labs like Zhipu and MiniMax will have little room to pivot toward proprietary models, even if they prove more profitable. The next major test will come when China’s next-generation AI models debut, likely later this year.

If they follow the same open-weight playbook, the financial strain on labs will intensify, while cloud providers like Alibaba stand to gain even more market power.

This article is for informational purposes only and does not constitute financial advice. Please consult a licensed financial adviser before making investment decisions.