[News] Alibaba Unveils AI Chip Zhenwu V900 for 1Q27 Mass Production, Maps Out New Server CPUs for 3Q27

Alibaba is accelerating its push into AI infrastructure with new in-house chips. According to ijiwei, Alibaba’s T-Head has unveiled the Zhenwu V900, an AI chip designed for both training and inference. CEO Eddie Wu said it is currently China’s most powerful AI chip, delivering three times the computing performance of the M890. As noted by EE Times China, the Zhenwu V900 is scheduled to enter mass production and begin large-scale deployment in 1Q27. T-Head also unveiled the timeline for its next-generation Zhenwu J900, which is set to launch in 3Q27.

The Zhenwu V900 features 216GB of memory and 1,200GB/s of chip-to-chip interconnect bandwidth, with native support for low-precision FP8 and FP4 computing, ijiwei notes. Beyond the chip itself, Alibaba is also targeting large-scale AI clusters built around the V900. EE Times China notes that T-Head’s in-house ICN Switch can connect up to 500,000 V900 chips in a single cluster, providing enough computing capacity to train and run frontier AI models with 5 trillion to 10 trillion parameters.

Alibaba is also preparing a new supernode system based on the V900. As noted by ITHome, the company plans to launch a new Panjiu supernode server powered by the Zhenwu V900 and ICN Switch in 1Q27. Meanwhile, Alibaba Cloud’s Lingjun Zhenwu M890 supernode, GP9A, is already available to customers, becoming the first supernode computing platform in China to successfully run a model with more than 2 trillion parameters.

T-Head Expands Server CPU Roadmap

Beyond AI accelerators, Alibaba is also pushing ahead with its server CPU roadmap. As noted by ITHome, the company unveiled plans for two new T-Head Yitian server CPUs for the first time, with both the Yitian 720 and Yitian 730 scheduled to launch in 3Q27. The report notes that the Yitian 720 is designed to improve single-core performance, core density, and energy efficiency. The Yitian 730 will take the roadmap a step further by introducing T-Head’s first fully in-house CPU microarchitecture.

As noted by ITHome, Alibaba’s T-Head plans to continue advancing its in-house CPU microarchitecture over the coming years. The upcoming Yitian 750, built on its second-generation in-house CPU core, will support T-Head’s proprietary ICN interconnect protocol, enabling direct connections between Yitian CPUs and Zhenwu AI chips to improve coordination between the two.

Alibaba Builds Out Full Data Center Chip Portfolio

With its AI accelerator and server CPU roadmaps expanding in parallel, Alibaba is building out a broader in-house chip stack for AI data centers. EE Times China notes that T-Head’s portfolio now spans four major categories: Zhenwu GPUs, Yitian CPUs, Panmai smart NICs, and ICN interconnect chips, covering all the core chip components needed to build hyperscale AI clusters.

Alongside the expansion of its in-house chip portfolio, Alibaba also outlined a long-term plan for its computing infrastructure. EE Times China notes that Alibaba plans to expand the combined capacity of its global data centers to more than 20GW by 2032. To support the expansion, the company has committed to investing more than US$53 billion, or around RMB 380 billion, in AI infrastructure over the next three years.

Read more

• [News] Alibaba T-Head Unveils Zhenwu M890 With 3× Performance; New AI Chips Planned for 3Q27/3Q28
• [News] Chinese Researchers Reportedly Push GAA With DUV, Opening Potential Path to 3nm-Level Performance

(Photo credit: Alibaba)