A tech news roundup says Alibaba paired the announcement with plans for 20-plus gigawatts of data-center capacity by 2032.
A tech news roundup shared on September 22, 2026, says Alibaba unveiled the Zhenwu V900 AI chip alongside plans for trillion-parameter Qwen models and 20-plus gigawatts of data-center capacity by 2032.
182.6K
18 Sources, first seen ago
—
Not ranked yet
—
Not ranked yet
Alibaba chip unit T-Head has unveiled the Zhenwu V900, a new AI accelerator the company says is intended for both training and inference. The announcement came at Alibaba's Apsara Conference in Hangzhou, where the company also described a broader system built around computing, networking and storage rather than a single chip.
T-Head says the V900 delivers three times the performance of its prior Zhenwu M890 chip. The company lists 216GB of memory, 1,200GB/s of inter-chip bandwidth, and support for lower-precision FP8 and FP4 computation. Those are company claims, not independently measured results.
The V900 is meant to connect through T-Head's ICN Switch interconnect chips. According to the company, more than 1,000 V900 chips can operate as one system, while a larger Alibaba Cloud cluster could scale to as many as 500,000 accelerators. The design reflects a familiar constraint of large-model computing: moving data and coordinating chips can become as consequential as raw processing speed.
T-Head says mass production and sales are planned for the first quarter of 2027. Alibaba is presenting the V900 as part of a full-stack infrastructure push that also includes server CPUs, smart network cards and storage controllers. The notable qualification is timing: the company has provided a roadmap and specifications, but the chip is not yet a broadly available product.
Sources: TechNode's conference report; Associated Press reporting.
Based on 163 sentiment-bearing replies from 152 accounts across 11 conversations.
Based on 163 sentiment-bearing replies from 152 accounts across 11 conversations.
