Huawei
Next-gen Huawei Ascend chip could debut in Q2 2026, rivalling Nvidia H100
Huawei is allegedly preparing several efficient Ascend AI chip products, and one of these could launch in Q2 2026. A new rumor has hit the surface, revealing the company’s AI chip strategies for the next year to compete with global chip tech rivals.
@rwang07 on X said that the next-gen Huawei AI chip – Ascend 910D could debut in Q2 2026. The new chipset may feature a four-die (multi-die) design this time.
A multi-die chip design provides several benefits over a single-die or dual-die layout. It not only improves the flexibility and performance of a chipset, but also its cost-effectiveness. It further optimizes process nodes and the overall chip creation.
Multi-die systems are also responsible for the integration of varied chiplets into a single package. Thus, the chip creation leads to various useful and effective capabilities.
Ray Wang said that the upcoming Huawei Ascend 910D can use a four-die configuration. It will be an upgraded version of the two-die layout used in the Ascend 910C.
Ascend 910C uses two separate silicon dies within a single package, which enables it to perform more complex functions and overcome limitations of single-die designs. The latest rumor now hints at better 910D performance with a quad-die setup.
Ascend 910D chip design and launch rumor image (Credit: rwang07/X)
The semiconductor analyst further said that the Ascend 910D AI chip might outperform the Nvidia H100. It may enter the commercial market by late Q2 2026.
Early reports said that the company started reaching major Chinese clients to analyze the technical value of the Ascend 910D. Some even said that the company sees the new chip as a powerful rival to Nvidia H100, and may release the first batch of samples by late May. But it seems there have been some changes in the plan.
Apart from 910D, the Ascend 920 is also rumored to be in development. It might debut directly with a dual-die design and can bring better process refinements with an extensive architectural shift to enable interoperability with Nvidia’s stack.
Note that the company hasn’t confirmed any of these details. Thus, we have to wait for some more inputs to get clarifications on this matter.
