Alibaba is scaling both sides of the AI race: larger models and custom silicon. At its Apsara Conference on Sept. 22, the company unveiled the Zhenwu V900, a new processor for AI training and inference that Alibaba says delivers three times the performance of its Zhenwu M890, released in May 2026.
Alibaba also confirmed that Qwen 4 is in training and said Qwen 4.5 and Qwen 5 are projected to scale to between 5 trillion and 10 trillion parameters. Together, the announcements show Alibaba expanding across custom chips, AI models, and cloud infrastructure, although pricing, software support, and customer access for the V900 remain unresolved ahead of its planned 2027 release.
Alibaba details V900 and its Qwen roadmap
The V900 has 216 GB of memory and 1,200 GB/s of inter-chip bandwidth, with native FP8 and FP4 support. Alibaba’s full-stack AI roadmap schedules the processor for mass production and commercial release in the first quarter of 2027. Its three-times-performance figure compares the V900 with Alibaba’s own M890 rather than a competing accelerator.
The wider architecture combines the V900 with Alibaba-developed networking and storage components. Alibaba says it can scale to clusters of up to 500,000 accelerator cards, while existing Zhenwu chips already serve more than 650 customers. Large systems increasingly depend on more than accelerator performance alone; recent rack-scale AI platforms similarly combine processors, memory, networking, and software to improve throughput.
At that scale, networking can become a separate bottleneck alongside power, cooling, and accelerator supply. Alibaba has not said it currently operates a 500,000-V900 cluster.
Qwen 4 is currently in training, while Qwen 4.5 and Qwen 5 are projected to reach 5 trillion to 10 trillion parameters. Alibaba’s current Qwen3.8-Max model has 2.4 trillion parameters, a figure also cited in Associated Press reporting on the new roadmap. Parameter count alone does not establish model quality, inference speed, or operating cost.
Alibaba has not disclosed whether those future Qwen models will run specifically on the V900.
Pricing and compatibility remain open
Alibaba has not published V900 pricing, specific V900-backed Alibaba Cloud instance types, regional availability, or support terms. Its announcement also does not explain how easily workloads built around Nvidia-oriented drivers, libraries, containers, and orchestration tools could move to V900 infrastructure.
Performance comparisons remain limited. Alibaba’s three-times claim uses the M890 as its baseline rather than Nvidia or other accelerators under independently reproduced testing. Memory capacity and inter-chip bandwidth also do not establish real-world training throughput, inference latency, or cost.
The company is expanding the cloud footprint around its AI strategy. On Sept. 23, Alibaba Cloud announced new cloud regions planned for Türkiye, Finland, and the Netherlands, plus additional capacity in Malaysia, Germany, the United Arab Emirates, France, and Hong Kong.
Pricing, cloud access, software compatibility, and independent testing will determine how V900 compares with established accelerator platforms once commercial availability begins in 2027.
Want to learn more AI tips, tricks, and prompting techniques? Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →