Alibaba Group Holding has unveiled what it described as China's most powerful artificial intelligence chip and signalled plans to train an AI model with up to 10 trillion parameters, as senior executives doubled down on the company's pursuit of artificial superintelligence (ASI). The announcements were made at the company's flagship annual Apsara Conference in Hangzhou on Tuesday, 22 September 2026.

Full-Stack AI Vision

Alibaba group chairman Joe Tsai opened the multi-day event — which runs through Thursday — by reaffirming the company's commitment to building end-to-end AI infrastructure. 'Alibaba is firmly investing in building full-stack AI ... to let AI move from a technological breakthrough to value creation,' Tsai said during his opening remarks. The statement positions Alibaba as an integrated player spanning silicon, software, and cloud services, rather than a pure application-layer competitor.

Three Cornerstones of the Machine Intelligence Era

CEO Eddie Wu Yongming used his keynote address to lay out the company's strategic road map, identifying 'AI models, AI chips and the AI cloud' as the three cornerstones of what he called the machine intelligence era. Central to the chip announcement is the T-Head Zhenwu V900, the in-house accelerator that Alibaba is positioning as a domestic alternative to offerings from Nvidia and Huawei Technologies' Ascend series. The Zhenwu V900 is reportedly designed to support the extreme compute demands of frontier model training at the scale Alibaba is now targeting.

The 10-Trillion-Parameter Ambition

The teased 10-trillion-parameter model would dwarf most publicly known AI architectures and, if realised, would represent a significant leap beyond Alibaba's existing Qwen model family. According to the company, scaling to that parameter count requires both proprietary silicon and a massive expansion of cloud infrastructure. Alibaba Cloud has set a target of scaling its global data-centre capacity to more than 20 gigawatts by 2032, a figure that underscores the capital intensity of frontier AI development.

Why It Matters

The announcements arrive as China's technology sector navigates ongoing restrictions on advanced semiconductor imports from the US, making domestic chip development a strategic imperative rather than a commercial option. Alibaba's dual push — proprietary silicon plus a record-scale model — mirrors moves by US hyperscalers such as Google and Meta, which have similarly invested in custom AI accelerators to reduce dependence on third-party chip suppliers. The competitive backdrop in China is equally intense, with Huawei Technologies' Ascend chips and models from rivals including Baidu and DeepSeek all vying for enterprise and cloud workloads.

What's Next

Alibaba has not disclosed a timeline for the 10-trillion-parameter model's training completion or public availability. Investors and industry observers will be watching whether the Zhenwu V900's real-world performance can match the company's claims, and whether Alibaba Cloud's infrastructure expansion stays on track amid a global data-centre supply crunch. The trajectory of Alibaba's ASI ambitions — and how regulators in both China and the US respond — will be a defining storyline for the broader global AI race.