China’s AI War Enters Phase Two

Alibaba’s Qwen rebrand marks the end of user acquisition and the start of execution hell.

“Order me lunch.” Three words. In 2024, that meant tapping through Ele.me. In 2025, Alibaba wants you to say it to Qwen and let the system do the rest.

On November 17, 2025, Alibaba renamed Tongyi to Qwen. The new tagline says it outright: “Can Chat, Can Do Things.” The bigger move was the plan to fold Amap, Ele.me and Fliggy directly into the Qwen interface. Alibaba is trying to replace the app layer itself.

This shift validates the thesis from August: the fight was never about search. It was about who controls the next operating system of daily life. The Do Engine. But the environment has changed. The capital-burning phase is ending. A more difficult era begins. Winning now depends less on money and more on execution — and the execution bar is brutal.

The Spending Pullback

Phase one was pure capital. Billions burned on ads. In October, DataEye showed the reversal. Yuanbao cut spending around 20 percent month over month. Doubao by 15 percent. The collective spend since early 2025 is near ten billion RMB. The bonfire did its job. Users arrived. Now the platforms must figure out how to keep them.

User growth is flattening. Doubao leads with 157 million MAUs. DeepSeek follows with 143 million. Yuanbao is at 41.6 million. But overall sector growth slowed to 3.4 percent in Q3. Early adopters are already captured. The next wave is harder: pragmatic users with stable habits. They only switch if AI is meaningfully better than what they already use.

Churn shows how fluid the market still is. Around 40 percent of DeepSeek users who churned moved to Doubao. Loyalty remains soft. The window to lock in dominant share is open — but closing.

Then comes the monetization problem. Serving compute-heavy inference to 150 million users is expensive. Subscriptions are a dead end in China. Ads would poison the conversational experience. That leaves one viable path: take a cut of the transaction. Which is exactly why the three giants are starting to diverge.

Three Different Bets

All three say “Let AI do things.” None can build the same thing.