While most of the Western AI chip conversation centers on Nvidia’s H100, B200, and whatever comes next from AMD or Intel, Huawei has been quietly building a parallel roadmap. At the 2026 Connect Conference, the company announced plans to release two new Ascend 960 AI chips in 2027, a move that tells you a lot about where Huawei thinks AI infrastructure is heading and how badly it wants a seat at that table.
David Wang, Huawei’s Rotating Chairman, confirmed the company is putting serious resources into AI chip development as part of a broader push on AI infrastructure. The Ascend 960 lineup will slot into that strategy as the next generation of Huawei’s homegrown accelerators, following the Ascend 910 series that has already seen adoption across Chinese cloud providers and AI labs locked out of Nvidia hardware by U.S. export controls.
That export control context is everything here. Since 2022, American restrictions have blocked Chinese companies from buying Nvidia’s most capable chips. That created a vacuum, and Huawei has been working to fill it. The Ascend 910B became a de facto alternative for companies like Baidu and Alibaba that needed to keep training large models. The 960 series is the next step in that progression, and the 2027 timeline suggests Huawei is pacing its releases to stay competitive as model complexity keeps climbing.
Details on specs are still limited at this stage. Huawei has not released benchmarks or technical specifics for the Ascend 960, so direct comparisons to Nvidia’s Blackwell architecture or Google’s TPU v5 are premature. But the existence of two variants is worth watching. It likely signals a split between a high-end training chip and a more inference-focused option, which is the same strategy Nvidia uses across its data center product line.
Pricing and availability details are also absent for now. Given that Huawei’s Ascend chips are primarily distributed through domestic Chinese channels and tied to Huawei Cloud, widespread global availability is unlikely. The real market here is China’s massive AI sector, which has enormous compute demand and limited legal access to leading Western silicon.
So why does this matter outside of China? A few reasons. First, if Huawei’s chips prove capable enough to train frontier-scale models, it reduces the leverage that U.S. export restrictions were designed to create. Second, it adds competitive pressure on the broader chip market, even indirectly. And third, it signals that AI chip development is now a multi-geography race, not a Silicon Valley monopoly.
The Ascend 960 is not a product that will appear in most developers’ cloud consoles anytime soon. But for anyone tracking the geopolitics of AI compute, or building infrastructure strategies that account for a fragmented global chip market, Huawei’s 2027 roadmap is a data point worth keeping in mind. The gap between Huawei and Nvidia is still real. But it is narrowing, and the pace of that narrowing is what makes announcements like this worth paying attention to.
