According to Wccftech, Bill Dally, chief scientist of GPU giant NVIDIA, pointed out that Huawei is tapping talents from NVIDIA's business in China, which means that Chinese competitors are catching up and posing an increasingly severe challenge to NVIDIA. This talent loss, coupled with restrictions imposed by the United States, has made Nvidia's position in the Chinese market vulnerable.
Bill Dally According to media reports, the US ban on H20 AI chips has provided Chinese companies with room for growth and helped them seize high-end AI talents, which is not good news for Nvidia. Dally said Huawei has successfully formed a huge AI research team composed of former Nvidia engineers, these engineers are currently developing software for the Chinese company.
Talent flow is worthy of attention, because Nvidia's strong position in the Chinese market is largely attributed to its robust software ecosystem CUDA. CUDA is considered irreplaceable and has always been the basis of AI computing architecture for large Chinese technology companies. Although Huawei has also developed its own alternative, CANN, it is said that its performance is still far behind CUDA and has not reached the same level of performance. This means that Huawei is trying to make up for its gap in the software ecosystem by attracting senior Nvidia engineers, especially in the field of software development, and further accelerates its competitiveness in the field of AI.
The report emphasized that Nvidia CEO Huang Renxun has always expressed opposition to the sanctions imposed by the Trump administration. These sanctions, especially the ban on H20 AI chips, have indeed created room for development for Chinese companies. Dally pointed out that in 2019, China's AI researchers accounted for less than one-third of the world's total, but now this number has grown to 50%, reflecting China's rapid development in the AI field and a significant increase in talent reserves.
Although Dally claims that Huawei's hardware technology is not as good as Nvidia, the situation is different when it comes to Nvidia's ability to ship the Chinese market. Due to US restrictions, Nvidia had to modify its products in the Chinese market, which led to Huawei's Ascend 910C and 910B chips becoming more suitable choices for domestic Chinese companies. The reason is that they are more easily accessible and are not affected by geopolitical uncertainty. This change in market dynamics provides Huawei with an opportunity to gain market share, which is a situation that Huang Renxun doesn't like at all. Relevant reports also pointed out that Nvidia warned that if restrictions persist, it may be difficult to compete in the Chinese AI market, which increases the risk of business termination.
Huang Renxun previously confirmed that Huawei's CloudMatrix AI cluster competes with Nvidia's Grace Blackwell series of products and said that nothing can stop Huawei. This means that Nvidia realizes that Huawei is rising and no longer regards it as an opponent that can be ignored. The article "Not Intel or AMD, and Huawei is the next opponent that Jensen cannot ignore, ready to take over Nvidia's AI market in China" also mentions Huawei's growing threat.
Intensified competition is not only hardware products, but also reflected in software and talents. Huawei poaches Nvidia engineers to strengthen software development capabilities, especially CANN efforts to create an alternative that can compete with the CUDA ecosystem. Although CANN's performance may still be inferior to CUDA at present, the addition of talents and continuous R&D investment will predict potential room for improvement.
