Huawei Technologies will release two new artificial intelligence chips in 2027, sharpening its challenge to Nvidia in the global AI computing market. The company's rotating chairman, David Wang, announced the Ascend 960DT for the first quarter of 2027 and the Ascend 960PR for the third quarter.
The strategy centers on UnifiedBus, Huawei's proprietary interconnect technology designed to connect thousands of processors for large-scale AI model training. Wang said the technology will "connect large numbers of chips to work together," positioning Huawei's pivot away from isolated chip specs toward holistic system performance.
This mirrors Nvidia's own moat: NVLink, not individual H100 or H200 specifications, is what locks customers into the Nvidia ecosystem. Efficient interconnects determine real-world throughput and operational costs in large language model training, where thousands of chips run in parallel. Data communication speed between processors directly impacts a data center's economics.
U.S. sanctions have restricted Chinese enterprises' access to Nvidia's latest accelerators, creating strategic necessity for Huawei. The company has already shipped over 1,000 AI computing systems to more than 370 customers—real deployments, not pilots, indicating substantial customer capital allocation despite export controls.
The 960DT likely targets training workloads requiring intensive computation, while the 960PR appears optimized for inference—deployment scenarios where lower power consumption and predictable latency matter more than raw throughput. This workload specialization reflects how enterprise AI customers actually buy infrastructure.
Huawei builds on its existing Ascend 910 series, which powers AI clusters in Chinese data centers. The company has navigated U.S. export restrictions by increasing reliance on domestic fabrication partners and advanced packaging techniques to sustain its roadmap.
The customer base—major cloud providers, research institutions, and state-linked enterprises—gives Huawei a critical feedback loop for software refinement. Its CANN toolkit directly competes with Nvidia's CUDA platform, which dominates global AI development. Software lock-in, not chip performance alone, determines platform longevity and revenue.
Huawei's 2027 roadmap reflects a recognition that system integration, not isolated chip specs, determines competitive advantage in AI infrastructure. The company is building an end-to-end stack to challenge Nvidia's ecosystem dominance in China.
