Alibaba used its Apsara conference in Hangzhou on Tuesday to deepen a full-stack AI push—unveiling the T-Head Zhenwu V900 processor it called China’s most powerful AI chip and outlining Qwen models that could scale to 5–10 trillion parameters—sending Hong Kong-listed shares up about 5.1%.

CEO Eddie Wu said the Zhenwu V900 delivers triple the performance of the prior M890 and can be clustered up to 500,000 cards to train and run the largest models, with mass production and commercial release targeted for the first quarter of 2027. Alibaba launched the M890 in May; Wu projected significant growth in annual AI chip shipments as Chinese firms race for domestic alternatives under tightening U.S. export curbs on Nvidia-class silicon.

On models, Wu said the Qwen team plans systems with 5–10 trillion parameters for longer-horizon tasks in pursuit of artificial superintelligence. Current flagship Qwen 3.8 Max has 2.4 trillion parameters; Alibaba said it is training Qwen 4, with 4.5 and 5 expected to reach the multi-trillion scale. Wu claimed “meaningful” progress on models that improve themselves with limited human involvement. He also set a goal for Alibaba Cloud global data-center capacity above 20 gigawatts by 2032, warning that mid-to-long-term AI demand still outpaces supply even as customer demand accelerates cloud revenue. AI supernodes are due online at commercial scale this quarter.

Wu framed the moment as comparable to the Industrial Revolution, predicting machines could eventually produce more than 1,000 times humanity’s “thinking,” versus under 3% today, while saying truly groundbreaking products have not yet arrived. Company and wire coverage agree on the chip claims, parameter roadmap, and share reaction; independent verification of performance versus Nvidia remains constrained by export controls and limited third-party benchmarks in the cited reports.