Huawei Reveals AI Chipmaking Roadmap and New Computing Power Plans

Huawei announced on Thursday that it will release four new versions of its Ascend AI chip over the next three years. This marks the first time the Chinese tech giant has openly shared details of its chipmaking ambitions, positioning itself to compete with Nvidia.

Huawei’s Push for Semiconductor Independence

The company has long been at the forefront of China’s efforts to build a domestic semiconductor industry. Its goal is to reduce dependence on a supply chain heavily controlled by the United States.

Huawei Vice Chairman Eric Xu confirmed that after launching the Ascend 910C earlier this year, the company plans to roll out two versions of its successor, the Ascend 950, in 2026. This will be followed by the Ascend 960 in 2027 and the Ascend 970 in 2028.

New Chips and Breakthroughs

Xu highlighted that computing power is central to artificial intelligence, especially for China’s AI ambitions. He revealed that the upcoming Ascend 950 will feature Huawei’s proprietary high-bandwidth memory. This development overcomes a long-standing bottleneck, previously reliant on suppliers from South Korea and the U.S.

Atlas Supernodes: Expanding Computing Power

Alongside the chips, Huawei will introduce new supercomputing nodes: the Atlas 950 and Atlas 960. These systems will support 8,192 and 15,488 Ascend chips respectively, making them the most powerful of their kind.

They follow the Atlas 900, also known as the CloudMatrix 384, which runs on 384 of Huawei’s latest 910C chips. According to research group SemiAnalysis, Huawei’s new products outperform Nvidia’s GB200 NVL72 system in certain benchmarks.

Huawei says its advantage comes from its “supernode” architecture, enabling ultra-fast interconnection between thousands of chips.