Huawei Expands Its AI Chip Roadmap
The two chips form part of Huawei’s broader effort to strengthen its AI computing business, building on the Ascend family as a domestic alternative for Chinese customers facing tighter access to advanced foreign semiconductors.
Huawei also outlined a longer roadmap, with the Ascend 970 planned for 2028 and the Ascend 980 for 2029. The company says its “Tao’s Law” approach aims to continue doubling compute specifications through architectural innovation. Huawei has not provided enough independent performance data to establish how the new processors will compare with Nvidia’s latest products, leaving their commercial impact uncertain. The faster release schedule gives Chinese technology companies a clearer pipeline for future AI hardware as domestic demand for computing capacity grows.
Scaling Beyond Individual Chips
Huawei is also focusing on how its processors work together. Its UnifiedBus technology is designed to allow large numbers of AI chips to communicate and operate as a single system for increasingly complex workloads.
The company has developed 11 semiconductors using UnifiedBus, according to Wang. Huawei says its largest linked systems, known as superclusters, can support as many as one million AI processors. That approach shifts the focus from individual processor performance to building large computing systems from Huawei’s own hardware.
Huawei Builds Its AI Computing Network
Huawei has shipped more than 1,000 smaller linked systems, called supernodes, to more than 370 customers. These systems combine multiple AI chips to work on a single task. The company has not disclosed the customers’ identities or the number of processors used in each system.
The reported deployments give Huawei an existing base for expanding its AI infrastructure, although the company has not provided enough information to assess the scale or performance of those systems.
Nvidia Remains the Benchmark
Huawei’s push comes as Washington’s restrictions continue to limit China’s access to some advanced computing chips and semiconductor manufacturing equipment, increasing the importance of domestic alternatives for Chinese technology companies and AI developers.
Nvidia’s position is supported not only by its processors but also by its developer ecosystem and software infrastructure. Huawei is working to build its own ecosystem, with Wang saying its AI chips now have 5,270 monthly active developers. Software compatibility and developer adoption are therefore important parts of Huawei’s challenge, as customers need an ecosystem that allows them to build, deploy and scale AI applications efficiently.
The Next Test Is Commercial Scale
Huawei’s 2027 launches will provide a clearer test of whether its roadmap can translate into sustained adoption. Its reported supernode shipments and growing developer base show that the company has moved beyond chip development into broader AI infrastructure, but wider use will depend on availability, system performance, software compatibility and reliability.
For Nvidia, Huawei’s progress adds another layer to an increasingly fragmented Chinese AI-chip market. For Huawei, the larger objective is to establish Ascend as part of a complete computing stack spanning processors, interconnect technology and developer support.
The key test will be whether Huawei can turn its expanding hardware roadmap and ecosystem into reliable, large-scale AI infrastructure that Chinese customers can deploy commercially.