Huawei accelerates Ascend chip production, but HBM remains bottleneck

According to recent estimates, Huawei is expected to produce millions of Ascend chips this year, a notable jump in the Chinese effort toward computing self-sufficiency. The bottleneck is already identified: HBM (High Bandwidth Memory) is expected to limit the production rate as early as 2026. Huawei is building a vertically integrated manufacturing chain that now covers design, fabs, packaging and memory, yet the HBM supply chain remains dependent on foreign suppliers subject to U.S. export restrictions.

The United States controls more than 70 % of global FLOPs, and export policy is intended to preserve that gap. In response, China has cut supplies of rare minerals and magnets, materials critical to the semiconductor industry itself. Howard Lutnick said, “Renewing Nvidia GPU sales to China is required to renew the flow of those raw materials.” The situation is not a stable equilibrium; it is a power-based negotiation.

DeepSeek’s models continue to run on Western hardware, and training the next generation of the model has been postponed because of reliance on Huawei chips. The gap between the manufacturer’s promises and actual performance remains large, and it is not solely a software-optimization issue.

Huawei also created SiCarrier, a tool-company meant to replicate foreign manufacturing equipment, and has invested over $9 billion in acquiring machines for its own fabs and in reverse-engineering. SiCarrier recently raised $2.8 billion to build dedicated fabs that will be staffed by Huawei employees, aiming for full ownership of the logic, memory and packaging layers without dependence on TSMC or other Western equipment suppliers.

The hardware is described as “very impressive” by sources, but it is openly acknowledged to be less efficient than Western counterparts. The efficiency gap is measured in performance per watt, compute density and yield stability in high-volume production. When the goal is millions of chips, lower efficiency translates into higher electricity costs, larger fab footprints and longer time-to-market—three parameters Nvidia has already solved in earlier generations.

China’s strategy is clear: control every stack layer from silicon to token. Just as China blocked Google a decade ago to build a domestic internet industry, the United States now restricts AI hardware access, and China is responding by constructing a complete alternative. Huawei has been selected as the national champion to lead the effort, and its competition with Nvidia is no longer theoretical.