China’s AI Firms Scaled up on Open-Source Models. The Next Phase May Be Different

Why It Matters

The approach reshapes how Chinese tech giants capture value from AI, influencing global pricing, cloud competition, and the sustainability of open‑source ecosystems.

Key Takeaways

• Qwen reached ~1 billion downloads, the world’s top open‑source model.
• Alibaba plans to grow Model‑as‑a‑Service, now <10% of AI cloud revenue.
• Chinese firms use GPU‑rental inference to monetize free models.
• Pricing wars force startups like MiniMax and Zhipu to tighten licences.

Pulse Analysis

China’s AI landscape has embraced open‑source models as a growth engine, a tactic that contrasts sharply with the proprietary focus of many U.S. firms. Alibaba’s flagship Qwen series, now approaching a billion downloads, serves as a showcase for how free model distribution can create a massive user base and lock‑in developers. By lowering entry barriers, these models accelerate talent development in cash‑strapped regions, reinforcing China’s ambition to become a global AI hub while simultaneously feeding the demand for cloud‑based inference services.

Monetisation in this ecosystem hinges on the inference layer rather than the model itself. Companies rent out GPU capacity, charge for premium features such as security and optimisation, and increasingly adopt a Model‑as‑a‑Service (MaaS) pricing model tied to token consumption. Alibaba’s cloud arm, which commands roughly 36% of China’s market, still derives the bulk of AI revenue from infrastructure, but its strategic push to expand MaaS could shift a larger share of earnings away from low‑margin GPU leasing. Compared with U.S. players that enjoy 40‑50% margins on closed‑source models, Chinese firms operate on a volume‑driven, single‑digit margin framework, prompting them to seek incremental revenue streams.

The sustainability of this approach faces headwinds. Intense competition has triggered price wars, eroding profitability for startups like MiniMax and Zhipu despite soaring revenues. Recent licensing tweaks—such as MiniMax’s restriction on commercial use—signal a move toward hybrid open‑source strategies that protect revenue while preserving community goodwill. Tools like the OpenClaw AI‑agent amplify demand for inference, offering a potential pathway to higher pricing power. Yet, limited GPU supply and the looming release of DeepSeek’s next model keep the market in flux, making the balance between openness and monetisation a critical watchpoint for investors and policymakers alike.