The future of Nvidia in China is becoming increasingly uncertain. At the start of October 2024, the Chinese administration issued a recommendation to artificial intelligence (AI) companies, urging them to prioritize the use of chips produced in China . Fast forward ten months, and this recommendation has evolved into a requirement . The Chinese government is now mandating that state-owned data centers utilize at least 50% of Chinese integrated circuits in their servers.
In contrast, Nvidia has finally acquired the export license necessary to sell its H20 AI GPU . However, the Chinese government has effectively vetoed this chip. The Cyberspace Administration of China , which regulates internet activity within the country, is currently investigating the H20 GPU due to suspicions that it could contain a backdoor that might be exploited by foreign entities. This investigation severely limits the likelihood of China adopting this chip for its needs.
The direct fallout from this unfavorable scenario for Nvidia is evident: in the last quarter, the company did not sell a single H20 GPU in China, as noted by Shaun Rein , an expert in the Chinese economy and founder of the China Market Research Group (CMR) based in Shanghai. While this statement holds some truth, it does have a catch. For a considerable part of last quarter, Nvidia lacked the export license required to deliver this chip to Chinese clients, though it now possesses it. With this license in hand, it could have shipped thousands of these GPUs in recent weeks.
China Has Alternatives Designed to Compete with Nvidia Chips
Despite aggressive efforts from the U.S. government to stifle competition, cutting-edge AI chips are still finding their way into China. These chips are primarily being imported through secondary markets and parallel imports from countries like India, Malaysia, and Singapore , where U.S. regulations hold limited sway. Moreover, developers creating significant AI models that utilize CUDA have discovered reliable avenues through the international second-hand market.
Cambricon Technologies is a rising star among companies specializing in AI chip design.
China currently boasts three significant alternatives to Nvidia. Among them, Cambricon Technologies —while not as well-known as Huawei or Moore Threads —is one of the companies with the highest growth potential in the design of AI GPUs. Recently, it secured approval from the Shanghai Stock Exchange to raise $560 million for the development of four chips destined for AI model training and inference, and to create an alternative to CUDA .
Another key player, Moore Threads , has developed several GPUs targeted for AI applications that compete with advanced offerings from Nvidia, AMD, and Huawei. Noteworthy products in its lineup include the MTT S4000 and MTT S3000 cards, which offer capabilities that rival established market leaders. Interestingly, the MTT S80 card, designed for gaming and content creation, claims to deliver a computation capacity of 14.4 TFLOPS in single-precision floating-point operations.
The third essential player in China’s AI chip sector is Huawei . Its flagship product, the Ascend 910D chip, is designed to surpass the performance of Nvidia’s H100 GPU . Huawei has also recently introduced the Ascend 920 chip, strategically aimed at filling the void that the Nvidia H20 GPU will leave in China’s market. This chip is slated for large-scale production in the latter half of 2025 , employing 6nm integration technology purportedly developed in collaboration with SMIC (Semiconductor Manufacturing International Corporation).
Image | Nvidia | Zhang Kaiyv
More information | Shaun Rein
In Xataka | The U.S. gives Huawei a tremendous opportunity to fill the gap left by Nvidia in China’s AI chip market.