September 17, 2026, (Inside AI) — Huawei Technologies will release two new artificial intelligence chips in 2027, the company's rotating chairman David Wang announced on Thursday, sharpening its challenge to Nvidia in the global AI computing market.
The Ascend 960DT is scheduled for the first quarter of 2027, followed by the Ascend 960PR in the third quarter. Wang framed the launches around UnifiedBus, a proprietary interconnect technology he described as central to Huawei's next generation of large-scale AI computing systems.
"UnifiedBus will be key to Huawei's next generation of large AI computing systems, helping connect large numbers of chips to work together." David Wang, rotating chairman, Huawei Technologies.
Wang also disclosed that Huawei has shipped more than 1,000 of these systems to over 370 customers, a figure that signals the company's growing footprint in AI infrastructure despite export controls.
The announcement matters because Huawei is the most credible challenger to Nvidia's dominance in AI accelerators, particularly in China, where U.S. sanctions have restricted access to advanced chips like the H100 and H200. By committing to a 2027 roadmap, Huawei is signaling that it intends to compete not just on chip design but on the system level, where interconnect and cluster architecture often determine real-world performance.
UnifiedBus appears to be Huawei's answer to Nvidia's NVLink, a high-speed interconnect that allows multiple GPUs to function as a single, more powerful unit. Training large language models requires thousands of chips working in parallel, and the efficiency of that communication can make or break a data center's throughput. Huawei's emphasis on UnifiedBus suggests it sees system integration as a differentiator, not just raw chip specs.
The 960DT and 960PR will likely target different workloads. The "DT" designation may indicate a variant optimized for training, while "PR" could signal a part tuned for inference or a specific performance profile. Huawei has not released detailed specifications, and Inside AI could not independently verify the performance claims or the exact positioning of each chip.
What is clear is that Huawei is building on the Ascend 910 series, which already powers clusters in Chinese data centers. The company has faced manufacturing constraints due to U.S. export restrictions, but it has increasingly relied on domestic fabrication partners and advanced packaging techniques to keep its roadmap alive.
The customer numbers Wang cited, over 370 clients and 1,000 systems shipped, offer a rare glimpse into adoption. These are not pilot projects. They represent real deployments, likely across cloud providers, research institutions, and state-linked enterprises in China. That scale gives Huawei a feedback loop to refine its software stack, particularly its CANN toolkit, which competes with Nvidia's CUDA ecosystem.
Huawei's push comes as China accelerates its self-sufficiency drive in semiconductors. The company's ability to ship AI systems at this volume, despite sanctions, underscores how the restrictions have inadvertently spurred domestic innovation. It also raises questions about the long-term effectiveness of export controls as a tool to slow China's AI progress.
For global AI companies, Huawei's 2027 roadmap is a reminder that the hardware landscape is not static. Nvidia remains dominant, but alternatives are emerging, and buyers in regions outside U.S. influence may increasingly weigh Huawei's offerings. The competition could eventually pressure prices and accelerate innovation, even if Huawei's chips remain largely confined to China for now.
Wang's remarks did not include details on pricing, performance benchmarks, or manufacturing partners. Huawei typically reveals such specifics closer to launch. The company is also expected to expand its AI cloud services and partner ecosystem around the new chips.
As 2027 approaches, the key question is whether Huawei can translate its system-level ambitions into chips that match Nvidia's latest generations. The 960DT and 960PR will be the first real test of that strategy at scale.
