Full Breakdown
Alibaba Unveils China’s “Most Powerful” AI Chip and Targets Trillion-Parameter Models
By Drooid · · How we work
Core Announcement (September 22, 2026)
At the Apsara Conference in Hangzhou, Alibaba Group CEO Eddie Wu announced the Zhenwu V900 AI accelerator and a roadmap to train a model with 5 to 10 trillion parameters. The V900 is described as China’s most powerful AI chip, delivering three times the performance of the Zhenwu M890. Wu also outlined plans to expand Alibaba Cloud’s global data-center capacity to more than 20 GW by 2032 to support larger models and growing demand.
Technical Highlights
- Chip performance: The V900 triples the compute throughput of the M890 and can be linked in clusters of up to 500,000 units for frontier model training and inference.
- Model size: The upcoming model will be two to four times larger than the current flagship Qwen 3.8 Max (2.4 trillion parameters). Alibaba’s Qwen team aims for the 5-10 trillion-parameter range to tackle more complex, longer-horizon tasks.
- Recursive self-improvement: Wu said the Qwen team is exploring a technique where models autonomously identify limitations, design experiments, and generate training data.
Infrastructure Expansion
Alibaba targets over 20 GW of global data-center capacity by 2032, roughly tenfold the pre-2022 level. Mass production of the V900 is scheduled for Q1 2027, with commercial release of AI supernodes the same quarter. The company has pledged more than US$53 billion to AI development over three years and raised about US$10.2 billion in a Hong Kong share placement to fund chips, data centres, and model research.
Market Reaction
Following the announcement, Alibaba’s Hong Kong-listed shares rose about 5 percent, reaching their highest level in a month, according to Reuters. Analysts noted the move positions Alibaba against Western rivals such as Nvidia, Meta, and Amazon, which are also scaling AI-compute capacity.
Official Statements & Responses
Eddie Wu framed the rollout as a response to “exponentially rising demand” for AI services and likened today’s AI coding to early electrification applications. Alibaba’s release reiterated the V900 performance claims and the 20 GW data-center target, emphasizing end-to-end co-optimization of chips, servers, networks, and models.
Criticism & Opposition
U.S. officials have accused Alibaba and other Chinese developers of “illicitly accessing” American AI models. Beijing rejected the allegations as unfounded and warned of reciprocal actions if Washington targets Chinese AI firms, highlighting the geopolitical tension around AI hardware and model development.
Conflicting Reports & Gaps
Sources uniformly report the V900’s mass-production timeline (Q1 2027) and the 20 GW data-center goal for 2032, leaving no direct contradictions. Public details on the geographic distribution of the new data centres remain sparse, and the exact timeline for commercial AI supernodes is described only as “this quarter” without a precise date.
What’s Next
- Q1 2027: Begin mass production of the Zhenwu V900 and launch commercial AI supernodes.
- 2027-2032: Scale Alibaba Cloud’s data-center capacity toward the 20 GW target, supporting 5-10 trillion-parameter models.
- Ongoing: Continue development of the Qwen 4 series and its successors while monitoring supply-chain constraints that could affect rollout speed.