Sources: Huawei has approached some clients about testing the technical feasibility of the Ascend 910D, its newest AI chip that it hopes will beat Nvidia's H100
Superpower rivalry over semiconductors heats up despite Washington's attempts to block Beijing
Context & Ripple Effects
Huawei’s Ascend program has progressed from earlier data-center chip launches to customer evaluations: Chinese internet companies were already testing the Ascend 910C, and Huawei recently planned 910C mass shipments to Chinese customers.
The 910D feasibility outreach therefore extends Huawei’s effort to make Ascend a credible domestic AI-compute option as Washington seeks to constrain Beijing’s access to advanced semiconductors.
First-order effects
- Prospective customers can benchmark the Ascend 910D’s technical viability before making deployment commitments, while Huawei gains feedback on whether its performance target is attainable in real workloads.
- Nvidia’s H100 becomes the explicit comparison point for Huawei’s next chip, raising the competitive bar beyond the China-specific products Nvidia has used in the market.
Second-order effects
- If tests are promising, Chinese AI buyers gain more leverage to diversify procurement away from a single foreign supplier; Nvidia may face greater pressure on product positioning and pricing in China.
- The move makes manufacturing execution more consequential for Huawei: prior reporting that it struggled to ramp Ascend server-chip production shows that technical validation alone would not establish supply reliability.
Third-order effects
- Repeated customer testing and shipment plans could turn China’s AI-chip market into a more durable second-source compute ecosystem, with software compatibility and dependable supply as important as peak-chip benchmarks.
- The broader U.S.–China semiconductor contest is likely to shift from limiting access to particular accelerators toward determining whether domestic alternatives can be validated, produced, and deployed at scale.
The trend: This is one data point in the localization of AI infrastructure, as Chinese buyers test domestic accelerators against Nvidia’s established hardware benchmark.