China's AI Revenue Only One-Tenth of the U.S., Monetization Gap Becomes Apparent
nashnova research
Six leading Chinese AI firms generate a combined $11 billion in annualized AI revenue — roughly 10% of OpenAI and Anthropic's combined $105 billion — exposing structural weaknesses in consumer conversion, API pricing, and enterprise deployment costs.
$11 billion vs. $105 billion — where does the gap come from?
Epoch AI estimates that Alibaba, ByteDance, Zhipu AI, Moonshot AI, DeepSeek, and MiniMax together generate about $11 billion in annualized AI revenue. OpenAI and Anthropic combined reach roughly $105 billion.
This means → China's top six AI firms, added together, amount to one-tenth of two American companies.
Analysis cited by Wallstreetcn points to three structural causes: low consumer subscription rates, fierce API price wars, and high enterprise-customization costs.
Hundreds of millions of users — why so little revenue?
Douban (ByteDance) has 380 million MAUs, Alibaba's Tongyi Qianwen 170 million, DeepSeek 130 million — large user bases that have not translated into matching paid revenue.
MiniMax tells the clearest story: average MAU of 27.6 million, but only 1.77 million paying users — a conversion rate of roughly 6%. Consumer gross margin sits at just 4.7%, while its enterprise platform earns 69.4%.
In plain terms = users are willing to try, not to pay. Living on consumer subscriptions alone is virtually impossible right now.
The more popular the open model, the less its maker earns?
After Zhipu AI released GLM-5.3 Flash weights, the model's share of calls on OpenRouter plunged from 88% to 22% in 20 days. Overall usage rose 17%, but Zhipu's own daily token volume fell 70%.
This means → third parties grabbed the weights and self-hosted. Usage accrued to everyone else; revenue accrued to no one at Zhipu.
This reflects the core paradox of open-weight distribution: model adoption and developer revenue can move in opposite directions.
Independent labs raise billions — but do they earn?
Moonshot AI raised over $1.2 billion across two rounds in early 2026. 36Kr reported its overseas revenue now exceeds domestic revenue, and a single 20-day stretch topped all of 2025 — yet sustained profitability remains unproven.
At Zhipu AI, 73.7% of 2025 revenue came from on-premise deployments (gross margin 48.8%), but by H1 2026 that share had dropped to 13.5% as the company pivoted to standardized cloud and API services.
API gross margin improved from 18.9% to 24.6% — a positive direction, but delivery and maintenance costs for large custom projects still weigh on margins.
Why can Alibaba and ByteDance actually monetize?
Alibaba's Q2 2026 AI cloud-computing and compute-services revenue hit roughly $7.1 billion, up 45% year-on-year. AI product revenue reached $1.82 billion, marking 12 consecutive quarters of triple-digit growth.
In plain terms = even if Tongyi Qianwen is priced near zero, it funnels users into Alibaba Cloud. The model is the lead generator; the cloud is the profit center.
ByteDance monetizes indirectly: AI-generated short dramas cut content costs, prompting producers to spend more on Douyin ads — industry estimates peg peak daily ad spending by AI-drama producers at roughly RMB 120 million (~$17.9 million). IDC data shows ByteDance's Volcengine held 49.5% of China's public-cloud LLM token volume in 2025; ByteDance reportedly raised its 2026 MaaS revenue target to $4 billion, up from about $221 million in 2025.
What is the path forward for independent labs?
The core tension is now plain: platform companies convert AI costs into cloud revenue and ad revenue, while independents face a double squeeze on scale and margins.
This means → without a cloud, advertising, or user ecosystem as a backstop, independent model companies relying on subscriptions and API fees have not yet closed the profitability loop.
Whether independents can achieve sustainable profits without leaning on a major platform's ecosystem is the key test for the sector's entire valuation thesis.
市场有风险，内容仅供研究参考，不构成投资建议。