Alibaba has finally shown its cards: the next-generation Qwen model has already been trained on its own chips.

At Thursday evening’s earnings call, Alibaba CEO Eddie Wu for the first time laid out a comprehensive timeline for T-Head’s third-generation chips: Zhenwu M890 will be fully deployed this year, V900 will enter mass production in the first quarter of 2027 with three times the computing power of the M890, and J900 will follow in 2028. Qwen’s next-generation flagship model is already being trained on T-Head chips, and Bailian’s inference services have also switched to domestically made chips.

The numbers add up, too: Alibaba’s AI-related product revenue this quarter was 17.9 billion yuan, accounting for 49% of the cloud business’s external recurring revenue, with annualized revenue already exceeding $10 billion. Overall cloud revenue was 61 billion yuan, up 53% year over year.

I think the key here isn’t the chip specs, but the timeline: V900 entering mass production in 2027 and J900 in 2028. Domestic computing power is moving from “usable” to “sufficient,” and Alibaba is the first to put its complete roadmap on the table. In the short term, domestic computing power will remain a recurring market theme; in the long term, players who control both their models and chips will find the cost equation increasingly favorable.
#AI Frontier
$BTC