Nvidia’s survival in China is becoming increasingly complicated. As we discussed over the past two weeks, the Cyberspace Administration of China (CAC) is thoroughly investigating Nvidia’s H20 GPU. This regulatory body governs internet activity in China, overseeing technology companies and ensuring compliance with laws regarding data security and personal information protection.
The challenge facing Nvidia is significant, as the CAC suspects that the H20 chip might contain a backdoor that could be exploited. If these allegations are true, it could jeopardize the chip’s use within the country. Meanwhile, growing calls from various sectors in China are urging local clients to refrain from purchasing the H20 GPU due to these prospective risks. These voices are notably influential.
Recently, a media outlet affiliated with the Central Chinese State Television criticized Nvidia fiercely, stating, “When a chip is not ecological, advanced, or secure, consumers have the option of not buying it.” In an effort to mitigate the rising concerns, Nvidia’s Security Director , David Reber Jr., published a blog post underscoring that the H20 chips contain “no backdoors, no kill switches, and no spyware.” This was a clear attempt to defend the company’s reputation amidst the increasing skepticism in China.
All Against Nvidia
In response to the investigations, the Chinese government is promoting an initiative encouraging local companies to develop AI models using homegrown chips . This movement has already gained traction. Leading firms such as Stepfun (part of Tencent Holdings) , Infinigence AI , Siliconflow (a Huawei subsidiary) , and several others have formed an alliance aimed at diminishing Nvidia’s foothold in the Chinese market.
The alliance for innovation in the model-chips ecosystem seeks to stop Nvidia in China
The collaboration, termed the Alliance for Innovation in the Model-Chips Ecosystem , marks a crucial step forward in this endeavor. Although Nvidia has the advantage of widespread adoption of its Compute Unified Device Architecture (CUDA) technology in ongoing AI projects, the landscape is gradually shifting. This framework provides developers with the necessary tools to build software optimized for Nvidia GPUs, but a transition to alternative technologies poses challenges for current projects.
Alongside this, Huawei is eager to capture a significant portion of the Chinese AI chip market with its Compute Architecture for Neural Networks (CANN) , which serves as an alternative to CUDA. However, Huawei is not the only player. Moore Threads , founded in 2020 under the leadership of Zhang Jianzhong, a former Nvidia executive, aims to provide compatible hardware and software solutions that eschew dependence on U.S. technology.
Moore Threads has introduced several GPUs for AI applications that theoretically compete with higher-end offerings from Nvidia, AMD, and Huawei. Furthermore, the company has rolled out a software suite called Muse , which is designed to replace CUDA, facilitating code portability from the Nvidia ecosystem to its own MTT cards. This feature is particularly attractive for those looking to transition from CUDA while minimizing disruption.
With these developments unfolding, the stakes are exceedingly high for Nvidia. If it cannot assuage the concerns of Chinese regulators and consumers, it risks losing significant market share in one of the world’s largest tech hubs. The situation underscores a potential paradigm shift in the global tech landscape, where local talent and innovation may drive future advancements in AI and computing.
In summary, Nvidia’s operations in China are facing unprecedented challenges. Regulatory scrutiny, combined with competitive pressure from local companies, places the chip manufacturer in a precarious position. The outcome of this tug-of-war will not only affect Nvidia but could also have wider implications for international tech relations.