While DeepSeek’s efficient AI model has fueled debate on NVIDIA’s dominance, China’s Huawei also puts pressure on the U.S. chip giant with its Ascend 910C, a variant of the 2019 Ascend 910 AI chip. According to the reports from Wccftech and Tom’s Hardware, citing Chinese media outlet AGI Hunt, Ascend 910C achieves 60% of NVIDIA H100’s performance, delivering strong inference results.
Interestingly enough, the breakthrough was released by DeepSeek’s researchers after they tested Huawei’s chips, the reports add. Anyhow, the information demonstrates China’s AI progress under U.S. restrictions, as the country gradually reduces its reliance on NVIDIA’s GPUs.
The reports suggest that Huawei’s Ascend 910C is a fully in-house chip built on SMIC’s 7nm N+2 process, featuring 53 billion transistors. Like its predecessor, the Ascend 910, it uses chiplet packaging, but while the original was made by TSMC on N7+, the 910C’s compute chiplet is produced by SMIC.
On the other hand, the Tom’s Hardware report also points out that Huawei’s Ascend 910C is not the best choice for AI training, where NVIDIA remains the leader of the domain.
Citing DeepSeek’s Yuchen Jin, the AGI Hunt report also notes that Chinese chips still struggle with long-term training reliability due to NVIDIA’s deeply integrated ecosystem, CUDA, which has been built over two decades. While inference can be optimized, Huawei still needs to improve its hardware and software for sustained training workloads, the report adds.
Meanwhile, this is not the only challenge NVIDIA has to tackle recently. According to Bloomberg, the White House and the FBI have launched an investigation into whether China’s AI startup DeepSeek illegally obtained NVIDIA chips through third-party buyers in Singapore and other locations.
Furthermore, US President Donald Trump remains firm on tariffs, despite meeting with NVIDIA CEO Jensen Huang. Speaking to the press on Friday last week, he confirmed plans to impose tariffs on chips, oil, and gas, likely by February 18, as noted by The Register.
Read more
(Photo credit: Huawei)