Alibaba Unveils China’s Most Powerful AI Chip, Teases 10-Trillion-Parameter Model
CEO Eddie Wu laid out a full-stack AI roadmap spanning chips, models and data centers, doubling down on artificial superintelligence just as Western AI leaders are calling for the opposite.
Alibaba Group CEO Eddie Wu used the company’s annual Apsara Conference in Hangzhou on Tuesday to detail a sweeping AI expansion across three fronts.
The company introduced its Zhenwu V900, calling it China’s fastest chip, teased an AI model as large as 10 trillion parameters, and set a 2032 target of more than 20 gigawatts of global data center capacity.
SCMP notes, chairman Joe Tsai framed the push as building “full-stack AI” to move the technology from breakthrough to value creation. Meanwhile, Wu described the moment as the dawn of a “Machine Intelligence” era he compared to the Industrial Revolution.
A Chip Built for China’s Chip Shortage
As Reuters details, the Zhenwu V900, developed by Alibaba’s T-Head semiconductor unit, delivers three times the performance of its predecessor, the M890 released in May, Wu said.
A single cluster can support up to 500,000 cards for AI training and inference, with mass production and commercial release planned for the first quarter of 2027.
Earlier reports note that T-Head has shipped more than 560,000 Zhenwu chips to over 400 customers across 20-plus industries.
This broad footprint strengthens Alibaba’s position against domestic rivals like Huawei in overcoming U.S. chip sanctions, even as reports suggest startup Moonshot relied on Alibaba’s Nvidia clusters to train its Kimi AI.
Wu was candid about the strain: he said industry demand for AI infrastructure “far outpaces” Alibaba’s supply capabilities, even as the company prepares to bring AI supernodes online at commercial scale this quarter.
Qwen’s Next Leap and a Quiet Admission
Alibaba’s current flagship Qwen model runs on roughly 2.4 trillion parameters; the planned successor, Wu said, would scale to between 5 trillion and 10 trillion.
In a separate statement, Alibaba said Qwen 4 is already in training, with Qwen 4.5 and Qwen 5 to follow.
Wu reportedly said that Qwen made meaningful progress in recursive self-improvement, where models spot limitations and generate data to evolve autonomously, fueling Alibaba’s push for artificial superintelligence.
According to the company’s press room, machines currently produce less than 3% of humanity’s total “thinking,” predicting that figure will eventually exceed 1,000%.
To fund the buildout, Alibaba has committed more than $53 billion to AI infrastructure over three years per SCMP, following an August share sale that raised roughly $10.2 billion in Hong Kong.
Alibaba Is Sprinting Where Others Want to Slow Down
The timing here is hard to ignore.
Wu’s comments about recursive self-improvement land barely a week after Anthropic’s Dario Amodei, joined publicly by Sam Altman and Elon Musk, called for a coordinated industry slowdown specifically because self-improving models could outrun human oversight.
Alibaba is announcing progress on precisely that capability, at precisely the moment its Western counterparts are proposing to pump the brakes.
That divergence isn’t accidental.
A coordinated slowdown only works if every major lab signs on, and Alibaba has never been part of that conversation.
U.S. export controls have already pushed it toward self-sufficiency, so slowing down carries none of the strategic logic it might hold for a company still buying Nvidia chips.
Source: Alibaba deepens AI push with new chip, bigger model; shares jump 5%
