The smartphone displays the flags of China and the United States, with an NVIDIA chip visible in the background.
Raa | Noor Photos | Getty Images
As Beijing continues its efforts to wean itself off reliance on U.S. technology, Chinese companies are stepping up efforts to produce viable alternatives to Nvidia's chips that power artificial intelligence.
U.S. sanctions against China in the past few years NVIDIAAnalysts told CNBC that China's dominance in the field poses a huge challenge to Beijing's efforts, at least in the short term.
Nvidia's boom has been driven by large cloud computing vendors buying into its server products that include its graphics processing units (GPUs). These chips enable companies such as ChatGPT maker OpenAI to train their vast artificial intelligence models on vast amounts of data.
These AI models are the foundation for chatbots and other emerging AI applications.
The U.S. government has restricted the export of Nvidia's most advanced chips to China starting in 2022, and restrictions were tightened last year.
Such semiconductors are key to China’s ambitions to become a leader in artificial intelligence.
CNBC spoke with analysts who identified some of the leading Chinese competitors looking to challenge Nvidia, including tech giant Huawei; Alibaba and Baidu and startups such as Biren Technology and Enflame.
The overall view is that they are currently behind Nvidia.
“These companies have made significant progress in developing application-specific artificial intelligence chips (ASICs),” Counterpoint Research senior analyst Wei Sun told CNBC.
“However, competing with Nvidia still faces significant challenges in terms of technology gaps, especially in general-purpose GPUs. Matching Nvidia is unlikely in the short term.”
Main challenges facing China
Sun said Chinese companies “lacked technical expertise,” underscoring one of the challenges.
However, U.S. sanctions and their knock-on effects are the biggest obstacle to China's ambitions.
Some of China's leading Nvidia challengers have been placed on the U.S. Entity List, a blacklist that restricts their access to U.S. technology. At the same time, a Multiple U.S. restrictions have restricted the export of key semiconductors and machinery related to artificial intelligence to China.
Chinese GPU manufacturers design chips and rely on manufacturing companies to produce them. For a while, this was supposed to be Taiwan Semiconductor Manufacturing Companyor TSMC. But U.S. restrictions mean many companies are unable to obtain chips produced by TSMC.
Therefore they must resort to SMICChina's largest chipmaker, its technology is still several generations behind TSMC. It lags in part because Washington restricts SMIC's access to key equipment from Dutch companies ASMLwhich is needed to make state-of-the-art chips.
Meanwhile, Paul Triolo, a partner at consulting firm Albright Stonebridge, said Huawei has been pushing to develop more advanced chips for its smartphones and artificial intelligence chips, which are taking up SMIC's production capacity.
“The key bottleneck will be domestic foundry leader SMIC, which will face the complex issue of how to allocate its advanced node production among Huawei, which currently holds the lion's share, GPU startups and many other Chinese design companies. of limited resources.
Nvidia is more than just GPUs
Nvidia's success is driven not only by its advanced semiconductors but also by its CUDA software platform, which allows developers to create applications that run on the U.S. chipmaker's hardware. This has led to the development of a so-called ecosystem around Nvidia's products, which may be difficult for others to replicate.
“That's the key, and it's not just about the hardware, it's about the entire ecosystem, the tools for developers, and the ability to continue to grow that ecosystem as the technology advances,” Triolo said.
Huawei leads
Triolo sees Huawei as one of the leaders in China's Ascend series of data center processors.
The company's current generation of chips is called the Ascend 910B, and the company is preparing to launch the Ascend 910C, which one person said could rival Nvidia's H100 product. The Wall Street Journal reported in August.
In its annual report earlier this year, Nvidia clearly regarded Huawei and other companies as competitors in fields such as chips, artificial intelligence software and network products.
Triolo said that Huawei “has many advantages” in terms of software and developer community building. But it faces similar challenges to the rest of the industry as it tries to compete with Nvidia.
“The GPU software support ecosystem is more entrenched around Nvidia, and to a lesser extent AMD, and Huawei faces significant challenges both in producing sufficient quantities of advanced GPUs (such as part of the Ascend 910C) and in continuing to innovate and improve the performance of GPUs. Given that the U.S. export controls limit SMIC’s ability to produce advanced semiconductors,” Triolo said.
Chip IPO ahead of schedule?
In the past two years, the challenges faced by Nvidia's Chinese competitors have been obvious. In 2022, Biren Technology conducted a round of layoffs, followed by Moore Threads the following year, with both companies blaming U.S. sanctions.
But startups are still holding out hope, hoping to raise funds to achieve their goals. Burundi It was reported last week that both Enflame and Biren were looking to go public to raise capital.
“Biren and other GPU startups are staffed by experienced industry personnel from Nvidia, AMD and other leading Western semiconductor companies, but they face the additional challenge of lacking the financial depth of Huawei,” Triolo said.
“As a result, both Biren and Enflame are pursuing initial public offerings in Hong Kong to raise funds for additional hiring and expansion.”