Huawei’s Ascend 910C Chip Matches NVIDIA’s H100 Yield Rates; Poised for the Next “DeepSeek Moment”

Reports are buzzing about Huawei’s impressive progress with its Ascend 910C chip, which has reached yield rates of up to 40%. This stride could soon translate into the production of over 100,000 units within the year, marking a significant development for the domestic AI industry.

Huawei is preparing to roll out its next-generation Ascend AI chip, and it’s looking like this release could shake up the current landscape, especially with companies like NVIDIA in the mix. The Ascend 910C is on the horizon, aiming to match NVIDIA’s powerhouse H100 AI GPU in terms of sheer performance. The Financial Times outlines that Huawei has significantly improved its chip yield rates over the last year, which means mass production might not be very far off.

According to insiders familiar with the situation, Huawei has pushed the “yield”—the ratio of working chips produced—to nearly 40%. This is a notable leap from the 20% yield they recorded around this time last year.

With such breakthroughs, it’s likely we’ll see the Ascend 910C hit the market sooner rather than later. Although some speculate that its debut might coincide with NVIDIA’s GTC 2025 to steal a bit of the spotlight, this remains unconfirmed. The enhanced yield rates set Huawei on a path to potentially significant profits, enabling them to enhance their presence in this segment.

Moreover, Huawei is reportedly gearing up for large-scale production of the 910C, planning to churn out up to 100,000 units by the end of 2025. With NVIDIA’s CUDA system currently dominating the Chinese market, Huawei is developing its own CANN software as a formidable competitor. It’s a long-term vision, but Huawei appears to be all in.

Currently, the Ascend AI chips are used by industry giants like DeepSeek, Alibaba, and Baidu, though they have mostly been reserved for low to mid-level tasks, particularly inference tasks. The advent of the 910C could broaden Huawei’s horizons, allowing it to tackle more demanding projects and perhaps even capture some of NVIDIA’s market share domestically.