China’s AI firms scaled on open‑source models. The next phase may be different

Open‑source gave Chinese AI a running start

Chinese AI companies built rapid scale by giving away the trains and charging for the stations. Alibaba Group Holding (阿里巴巴) made open‑sourcing central to its AI strategy with Qwen; it has been reported that Qwen has reached nearly 1 billion cumulative downloads over three years. At a public forum, Alibaba chair Joe Tsai famously shrugged that “we don’t make money from AI,” underscoring a wider industry playbook: seed ecosystems with free model weights, then monetise adjacent services.

From free models to paid services

That playbook worked. The wave that began with early releases such as DeepSeek helped domestic start‑ups — reportedly including MiniMax and Zhipu AI (智谱AI) — scale swiftly and even reach blockbuster listings in Hong Kong. But free models compress margins. Open‑source removes straightforward IP monetisation, so firms have leaned on cloud inference, paid APIs, enterprise fine‑tuning, and vertical applications to extract revenue. Incumbent cloud players and platform owners are turning openness into billable infrastructure.

A hybrid future shaped by markets and geopolitics

So what comes next? Expect a hybrid approach: continued open releases to maintain developer mindshare, paired with proprietary fine‑tuned models, safety/compliance layers, and enterprise solutions behind paywalls. Market pressure from investors and “thin margins” are pushing the shift. Geopolitics matters too — export controls, sanctions and tighter trade policy have raised the premium on software differentiation and on domestic supply chains, meaning Chinese firms may need to balance openness with strategic control. Openness bought them speed. Now, monetisation and resilience will test whether that generosity can be sustained.