Ascend 950 Super Node Launches in 2026 as KunTai Unveils AI Computing Matrix

On September 17, the Ascend AI Industry Summit was held, at which the Ascend 950 Super Node was officially launched. At the same time, KunTai, a subsidiary of Digital China, released a new-generation AI computing product matrix based on the Ascend 950 technology roadmap.

The product matrix covers the KunTai A989 I5 standard server, the KunTai A989 I5 air-cooled super node, and the KunTai PoD2000 A5 liquid-cooled super node, and is suited to core AI business scenarios such as large-model training and high-concurrency intelligent inference.

Among them, for large-scale data center construction, trillion-parameter large-model training, and central high-concurrency inference scenarios, a single KunTai PoD2000 A5 super node integrates 1,024 cards, supports 256TB unified memory addressing, and achieves a mean time between failures of 300 hours for 1,000 cards. Relying on TB-level NPU interconnect ultra-high bandwidth and 3μs ultra-low RTT latency, the product breaks through communication bottlenecks among computing, storage, and other resources within the cluster.