Alibaba's T-Head Unveils Yitian Server CPU Roadmap, Yitian 720 and 730 to Launch in 2027

en.Wedoany.com Reported - On September 22, at the 2026 Hangzhou Apsara Conference, Alibaba's T-Head publicly unveiled its Yitian server CPU product roadmap for the first time, planning to launch the Yitian 720 and Yitian 730 server CPUs in 2027, and to subsequently continue advancing the Yitian 750 based on second-generation self-developed CPU cores. The Yitian series will further coordinate with T-Head's Zhenwu AI chips, ICN interconnect chips, and other products to expand data center computing infrastructure for large model training, inference, and Agentic AI scenarios.

According to the roadmap announced this time, the Yitian 720 will be upgraded in three directions: single-core performance, core density, and energy efficiency. The Yitian 730 will, for the first time, adopt T-Head's fully self-developed CPU microarchitecture, with single-core SPECint2017/GHz performance reaching up to 1.4 times that of the Yitian 710, further enhancing the performance density of server nodes in general-purpose computing, data processing, and AI-assisted computing scenarios.

In subsequent product planning, T-Head will also launch the Yitian 750. This chip will be designed based on second-generation self-developed CPU cores and will support T-Head's self-developed ICN die-to-die interconnect bus protocol. Through the ICN bus, Yitian CPUs can achieve direct interconnection with Zhenwu series AI chips, shortening the data transmission path between CPUs and AI accelerator chips, and improving data exchange efficiency and coordination capabilities within computing nodes.

T-Head has now formed a chip product system covering multiple key links in data centers, including the Zhenwu series AI chips, Yitian series server CPUs, ICN Switch interconnect chips, Panmai series smart NICs, and Zhenyue series storage controller chips. At the concurrently held Apsara Conference, T-Head also released a new-generation training-inference integrated AI chip, the Zhenwu V900, whose single-chip performance reaches 3 times that of the Zhenwu M890, is equipped with 216GB of memory, achieves die-to-die interconnect bandwidth of 1200GB/s, and is planned to enter mass production in the first quarter of 2027.

Around the construction of large-scale AI clusters, Alibaba is advancing the collaborative design of CPUs, AI chips, interconnect chips, smart NICs, storage controllers, and server systems. The new-generation supernode servers equipped with the Zhenwu V900, ICN Switch, Panmai smart NICs, and Zhenyue SSD controller chips will improve cluster bandwidth and computing resource utilization through system-level interconnection, and support the expansion of a single AI cluster to larger scales. After the Yitian 750 joins the ICN interconnect system, system-level coordination between CPUs and AI chips will further extend toward high-speed chip-to-chip interconnection.