Chinese companies aiming to compete with Nvidia on AI chips

Long-standing U.S. sanctions on China’s semiconductor industry have forced Beijing to step up efforts to promote its domestic semiconductor industry.

The boom in artificial intelligence and basic models will only accelerate China’s goal of assuming a leading role in the semiconductor industry.

For now, American companies NVIDIA The company designs the core hardware needed to train huge AI models like those used by OpenAI, which powers ChatGPT, and its graphics processing units (GPUs) have been a hot topic.

While Nvidia can export certain chips to China, Washington has signaled its intention to cut off cutting-edge semiconductors and the tools needed to make them from its tech rival, drawing new attention to China’s efforts to build its own semiconductors to rival Nvidia and power the world’s second-largest economy’s AI industry.

CNBC spoke to two analysts who identified some of Nvidia’s main Chinese competitors.

Huawei is one of China’s largest technology companies, with operations ranging from telecommunications infrastructure to home appliances and cloud computing. Its chip design division is called HiSilicon.

The Shenzhen-based company designs its Ascend series of data center processors. Huawei then sells these chips as part of servers that are installed in data centers to train AI models. The company’s AI servers are sold under the brand name Atlas.

The company’s current generation of chips is called the Ascend 910B, and the company is preparing to launch the Ascend 910C, which could be on par with Nvidia’s H100 products, he said. The Wall Street Journal reported in August.

In its annual report earlier this year, Nvidia explicitly identified Huawei as a competitor, along with other companies, in areas including chips, software for AI and networking products.

“This isn’t just about hardware, it’s about the whole ecosystem, the developer tools and the ability to continue to evolve that ecosystem as technology advances. Huawei has a lot of advantages here and is building a software ecosystem around its Ascend series of data center processors,” Paul Triolo, associate partner at consulting firm Albright Stonebridge, told CNBC.

Alibaba Both Microsoft and Baidu buy Nvidia chips, but also design their own semiconductors for AI processes.

Baidu, one of China’s largest internet companies, designs its own chips for use in servers and self-driving cars, under the brand name “Kunlun.”

Alibaba’s semiconductor design division, T-Head, has developed an AI inference chip called Hanguang 800. Inference is the process that follows training an AI model and refers to the practical application of AI in the real world, such as a chatbot that responds to user questions.

“Alibaba’s AI inference chips have already been deployed to accelerate the recommendation system on its e-commerce platform, while Baidu has integrated its Kunlun chips into its data centers and autonomous driving units,” Wei Sun, senior analyst at Counterpoint Research, told CNBC.

Like Nvidia, Biren Technology designs general-purpose GPUs and has a software development platform for building applications on top of the hardware.

The chips are part of Biren’s Bili series of products designed for use in AI training in data centers.

Last year, Biren was added to a US blacklist known as the Entity List, restricting its access to certain US technology.

Cambricon Technologies designs many different kinds of semiconductors, from those designed for training AI models to those that can run AI applications on-device rather than in a data center.

However, the company continues to report large losses. Reportedly The company laid off employees last year, according to the South China Morning Post.

Cambricon Technologies is also listed on the US Entity List.

MooreThreads, founded in 2020, is developing GPUs.s It is designed for training large-scale AI models.

MTT KUAE is the company’s GPU-powered data center product. According to a statement on the company’s website, the company’s mission is to become the “global GPU leader.”

Big brands are also backing the app: TikTok’s owner, ByteDance, is an investor, along with other big venture capital firms such as Sequoia and GGV Capital.

Moore Threads is also listed on the U.S. Entity List.

Enflame Technology is another Chinese startup vying to become a homegrown alternative to Nvidia. The company designs chips for data centers with a focus on AI training and processing.

Tencent, one of China’s largest technology companies, has invested in Enflame.