News

China seeks a homegrown alternative to Nvidia — these are some of the companies to watch

Illustration of the China and U.S. flag on a central processing unit.
Blackdovfx | Istock | Getty Images
  • Nvidia with its graphics processing units, or GPUs, has garnered the headlines, as the key piece of hardware required to train up huge AI models, such as the likes seen from OpenAI.
  • With the U.S. attempting to restrict China's access to cutting-edge semiconductors, Beijing has ramped up efforts to nurture its homegrown chip industry.
  • Some of China's tech giants inclduing Huawei, Alibaba and Baidu are among those trying to create alternatives to Nvidia.

U.S. sanctions over the years on China's semiconductor industry has forced Beijing to ramp up efforts to boost its domestic chip sector.

The boom of artificial intelligence and foundational models has only spurred on China's goal of playing a leading role in the chip industry.

So far, it is American firm Nvidia with its graphics processing units, or GPUs, that has garnered the headlines, as it designs the key piece of hardware required to train up huge AI models, such as the likes seen from OpenAI that underpins ChatGPT.

While Nvidia can ship certain chips to China, Washington has shown its willingness to cut its tech rival off from the most cutting-edge semiconductors and tools needed to make them. This has renewed focus on China's homegrown efforts to rival Nvidia and create semiconductors that can underpin the world's second-largest economy's own AI industry.

CNBC spoke to two analysts who identified some of the leading Chinese competitors to Nvidia.

Huawei

Huawei is one of China's tech champions with a business that spans telecommunications infrastructure to consumer electronics and cloud computing. Its chip design unit is called HiSilicon.

The Shenzhen-headquartered company designs the Ascend series of data center processors. Huawei then sells these chips as a part of servers that go into data centers to train AI models. Its AI servers are under the brand name Atlas.

The firm's current generation of chip is called the Ascend 910B, and the company is gearing up to launch the Ascend 910C, which could be on par with Nvidia's H100 product, according to a Wall Street Journal report in August.

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

"It is not just about the hardware, but about the overall ecosystem, tools for developers, and the ability to continue to evolve this ecosystem going forward as the technology advances. Here, Huawei holds a lot of advantages, and is attempting to build a software ecosystem around its Ascend series of datacenter processors," Paul Triolo, a partner at consulting firm Albright Stonebridge, told CNBC.

Alibaba and Baidu

Alibaba and Baidu both buy Nvidia chips but they are also designing their own semiconductors for AI processes.

Baidu, one of China's biggest internet companies, designs its own chips for the use in servers and autonomous cars under the brand name Kunlun.

Alibaba's semiconductor design unit called T-Head, has developed an AI inference chip called the Hanguang 800. Inference is the process that follows the training of AI models, as it refers to the actual application of AI in the real world, such as a chatbot responding to user queries.

"Alibaba's AI inference chip has already been deployed to accelerate its recommendation system on its e-commerce platform. Baidu has integrated its Kunlun chip into its data centers and autonomous driving sector," Wei Sun, a senior analyst at Counterpoint Research, told CNBC.

Biren Technology

Like Nvidia, Biren Technology designs a general purpose GPU and has a software development platform to build applications on top of the hardware.

These chips form part of Biren's Bili series of products designed to be used in data centers for AI training.

Last year Biren was added to a U.S. blacklist known as the Entity List, which restricts its access to certain American technology.

Cambricon Technologies

Cambricon Technologies designs various types of semiconductors from those designed to train AI models to those that can run AI applications on devices, rather than in data centers.

However, the company has continued to report significant losses and reportedly laid off workers last year, according to the South China Morning Post.

Cambricon Technologies is also on the U.S. Entity List.

Moore Threads

Moore Threads, founded in 2020, is developing GPUs designed to train large AI models.

MTT KUAE is the company's data center product containing its GPUs. The company's mission is to become a "global GPU leader," according to a statement on its website.

It also has big brand names backing it. TikTok-owner ByteDance is an investor alongside big venture capital firms including Sequoia and GGV Capital.

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

Enflame Technology

Enflame Technology is another start-up in China vying to position itself as a homegrown alternative to Nvidia. The company designs chips for data centers focused on AI training and processes.

Tencent, one of China's biggest tech companies, is an investor in Enflame.

Clarification: Paul Triolo's title has been updated to reflect his latest position.

Copyright CNBC
Contact Us