Huawei announced plans on September 17 to release two enterprise processors, the Ascend 960DT and Ascend 960PR, setting a 2027 launch timeline aimed directly at competing with international artificial intelligence hardware makers.
Key points
- Huawei announced the Ascend 960DT for Q1 2027 and the Ascend 960PR for Q3 2027.
- The chips rely on Huawei's UnifiedBus interconnect technology to create high-density computing clusters.
- Huawei plans an annual release cycle, aiming for the Ascend 970 in 2028 and Ascend 980 in 2029.
- Export restrictions continue to block Nvidia from supplying top-tier enterprise AI accelerators to mainland China.
Speaking during a keynote presentation at the annual HUAWEI CONNECT event in Shanghai, Rotating Chairman David Wang confirmed the revised hardware schedule. The announcement outlines the company’s effort to address domestic demand for high-performance computing hardware while facing ongoing trade restrictions from the United States government.
Ascend 960DT and Ascend 960PR Launch Schedules
The Ascend 960 generation splits core data center workloads into two dedicated processors. The Ascend 960DT focuses on artificial intelligence model decoding and training tasks, while the Ascend 960PR handles prefill and inference processing.
Huawei adjusted its production roadmap to deliver both silicon platforms earlier than originally expected. According to Wang, the Ascend 960DT is scheduled for release in the first quarter of 2027, moving three quarters ahead of its earlier internal target. The Ascend 960PR is scheduled to follow in the third quarter of 2027, advanced by one quarter.
Huawei plans to launch the 960DT in the first quarter of 2027 and the Ascend 960PR in the third quarter.
The upcoming 2027 hardware builds on Huawei’s Ascend 950 series, comprising the 950PR and 950DT, which remains targeted for deployment across 2026.
Interconnect Technology and SuperCluster Architecture
Because trade rules limit access to leading global foundries, Huawei is structuring its high-performance roadmap around scale-up interconnect infrastructure rather than relying solely on raw single-die silicon metrics.
The company developed an 11-chip portfolio utilizing its proprietary UnifiedBus interconnect standard. This protocol connects individual processors into server racks known as SuperPoDs, including the Atlas 960E SuperPoD. These units incorporate optical networking systems, specifically the High-density Optical-interconnect-Node Engine (Hi-ONE), which uses near-packaged optics to minimize latency between chips.
SuperPoDs are designed to coordinate multiple NPUs through interconnect. We have developed a next-generation optical interconnect product based on near-packaged optics (NPO): the High
density Optical-interconnect-Node Engine (Hi-ONE).
Huawei reports that these systems can expand into modular computing networks containing up to 1 million interconnected neural processing units (NPUs). To date, the company has shipped more than 1,000 SuperPoD and supernode systems to over 370 enterprise customers across mainland China.
Domestic Momentum Against Industry Restrictions
Market conditions inside China provide a distinct backdrop for the Ascend roadmap. United States export controls prevent nvidia from shipping its highest-tier AI training accelerators, including its Hopper and Blackwell enterprise products, to Chinese organizations.
At the same time, Chinese regulatory authorities have encouraged domestic companies to reduce their dependence on imported semiconductor platforms. This dynamic creates a captive enterprise market for local manufacturers capable of providing operational alternatives for commercial model development and large-scale deployment.
Huawei has not released specific foundry partners, lithography nodes, memory configurations, or standard thermal design power ratings for either Ascend 960 processor. Benchmark evaluations conducted by independent third parties such as TechInsights have previously examined Huawei’s packaging and interconnect approaches across earlier Ascend revisions.
Roadmap for Ascend 970 and 980
Huawei confirmed it intends to maintain an aggressive cadence beyond the 2027 rollout. The company has structured its enterprise hardware group to work on a single-year replacement cycle, targeting double the performance with each generation.
We’re evolving our Ascend chip series on a one-generation-a-year cycle. In 2028 and 2029, we will roll out the Ascend 970 and 980 chips, respectively.
Standalone component pricing and retail accelerator board configurations for the Ascend 960 series have not been disclosed, as Huawei continues to distribute its flagship compute silicon primarily through full-rack enterprise installations.
