In October of the previous year, following the United States’ implementation of new regulations that blocked NVIDIA from selling cutting-edge artificial intelligence (AI) chips to China, NVIDIA quickly developed a downgraded version of these chips specifically tailored for the Chinese market to maintain its sales without violating the rules. However, due to significant performance differences between the downgraded and original chips—allegedly a reduction of 25% or more in performance—major Chinese cloud computing customers such as Alibaba, Tencent, Baidu, and ByteDance became less enthusiastic about purchasing them.
Sources with knowledge of the situation have stated that large Chinese cloud computing companies, including Alibaba and Tencent, have been testing NVIDIA’s China-specific chip samples for a while now. They have indicated to NVIDIA that their orders this year will be substantially less than initially planned.
Reportedly, many firms are also considering shifting orders for advanced semiconductors to domestic companies to reduce their reliance on NVIDIA. The uncertainty of NVIDIA chip supply due to policy restrictions poses a significant risk, and if these limitations become more stringent, companies with ambitions to progress in the AI cloud computing field may need to consider alternative solutions to mitigate potential risks related to chip restrictions in the future.
Alibaba’s Pingtouge Semiconductor previously released the Hanguang 800 AI chip.
On the other hand, the gap between domestic chips and NVIDIA’s is narrowing, which is a key reason why many AI heavyweights are turning their attention to local manufacturers. Various enterprises such as Alibaba’s Pingtouge and Moore Threads have showcased their chips or new developments, demonstrating advantages in performance, power consumption, and cost.
Indeed, as AI cloud services improve, more enterprises are experimenting with renting AI processing power and using cloud services as a workaround for NVIDIA’s China-specific chips and related computational power policy restrictions. With the growth of models like ChatGPT, the future direction is increasingly focused on AI applications instead of developing large language models. This approach is more cost-effective for many companies and represents a significant direction in AI application development. This shift could also be a contributing factor to the cooling interest in NVIDIA chips.
NVIDIA CEO Jensen Huang presenting AI Chips
However, some tech company insiders have mentioned that, given the broader ecosystem covered by NVIDIA’s products, there remains a distance to be covered by local alternatives, and NVIDIA’s chips will still be the preferred choice over the next year. For domestic AI chips to truly replace NVIDIA, there is still a gap in terms of computing power, energy efficiency, stability, and the software and hardware ecosystem. But the recent lukewarm response to NVIDIA in China suggests that this gap might be closing.
It’s indisputable that in the already-dawning AI era, competition for computational power will intensify. In light of China’s vast demand for computational power, restrictions on NVIDIA could potentially lead to more trade friction and technological competition in the future. The cold reception of NVIDIA chips in China reflects the complex rivalry between the U.S. and China in the AI domain. This struggle for computational power in the AI era is set to continue. China’s AI industry must accelerate its pace of independent innovation and cultivate its own AI chip brands to enhance computational power capabilities and maintain competitiveness and initiative in the AI age.