This is incredible, according to LatePost the 200 million users of ByteDance's AI chatbot Doubao (the most popular in China) barely generate $140K USD of revenue per day ... versus a $2B run rate for its video generation Seedance models which are primarily sold to businesses (at a 70% gross margin too)
And management made the decision to shift focus to enterprise after visiting Anthropic 2 months ago. (Really? I find that hard to believe, must be coincidental timing, no way that wasn't obvious earlier)
ByteDance's AI business is increasingly split between two very different stories.
Doubao has more than 200 million daily active users, making it one of the most widely used AI products in China. Yet daily revenue is reportedly less than RMB 1 million ($140K USD), with most of it coming from e-commerce commissions rather than direct AI monetization.
Seedance, ByteDance's AI video generation platform, looks very different. According to Chinese media reports, it has reached roughly RMB 1 billion in monthly revenue (about $140 million), equivalent to around $2 billion in annualized revenue, with gross margins reportedly near 70%. Nearly all of that revenue comes from enterprise customers.
The contrast suggests that the debate over consumer versus enterprise AI is becoming less theoretical.
According to reports, ByteDance leadership visited Anthropic and subsequently increased its focus on enterprise AI. The company expanded teams supporting coding models and reportedly set ambitious growth targets for its Model-as-a-Service business. Seedance has become the clearest example of that strategy. Video generation workloads can run efficiently on lower-cost domestic chips, avoiding some of the communication bottlenecks that make large language models expensive to serve. More importantly, customers are paying for a concrete output that directly reduces production costs.
Doubao highlights the challenge on the consumer side. Despite massive usage, monetization remains limited. Chinese users have grown accustomed to free digital services, and the rise of open-weight models such as DeepSeek has reinforced expectations that AI chat should be free. Unlike in the U.S., large-scale willingness to pay for AI subscriptions has yet to emerge.
ByteDance's response is to keep investing. The company is reportedly planning more than RMB 200 billion ($28 billion) of capex in 2026, roughly 60% of its estimated 2025 profit, while betting that enterprise products such as Seedance, coding tools, and MaaS offerings can grow into that infrastructure spend.
We've argued before that Chinese AI apps have been very successful at acquiring users but much less successful at monetizing them. Seedance's reported margins show that meaningful revenue will accrue in areas where AI delivers a measurable business outcome rather than a consumer convenience.
All revenue and margin figures come from LatePost. Reported margins also depend on assumptions around data-center depreciation, and recent reports suggest Seedance's growth may already be slowing. Even with those caveats, the contrast between Doubao and Seedance may be the clearest sign yet that AI value creation in China is shifting from consumer reach toward enterprise spending.