Chinese firms need to compete with Nvidia in AI chips

Years of US sanctions against China's semiconductor industry forced Beijing to step up its efforts to strengthen its domestic chip industry.

The boom in artificial intelligence and fundamental models has only further fueled China's goal of becoming a frontrunner within the chip industry.

So far it’s an American company NVIDIA with its graphics processing units (GPUs), which have made headlines for developing the important thing hardware needed to coach huge AI models, equivalent to those found at OpenAI on which ChatGPT relies.

While Nvidia can supply certain chips to China, Washington has shown a willingness to chop off its technology rival from probably the most advanced semiconductors and the tools needed to make them. This has put the main target back on China's domestic efforts to compete with Nvidia and develop semiconductors that may form the idea of the world's second-largest economy's own AI industry.

CNBC spoke to 2 analysts who identified a few of Nvidia's leading Chinese competitors.

Huawei is considered one of China's technology champions. Its business areas range from telecommunications infrastructure to consumer electronics and cloud computing. Its chip design department is known as HiSilicon.

The Shenzhen-based company develops the Ascend series of information center processors. Huawei then sells these chips as a part of servers utilized in data centers to coach AI models. Its AI servers run under the brand name Atlas.

The company's current generation of chips is known as the Ascend 910B, and the corporate is preparing to launch the Ascend 910C, which could compete with Nvidia's H100 product, based on an August Wall Street Journal report.

In its annual report earlier this 12 months, Nvidia specifically identified Huawei, amongst others, as a competitor in areas equivalent to chips, AI software and networking products.

“It's not just about the hardware, it's about the entire ecosystem, the tools for developers and the ability to evolve that ecosystem as technology advances. This is where Huawei has a lot of advantages and is trying to build a software ecosystem around its Ascend line of data center processors,” Paul Triolo, associate partner at consulting firm Albright Stonebridge, told CNBC.

Alibaba and Baidu each buy Nvidia chips, but additionally develop their very own semiconductors for AI processes.

Baidu, considered one of China's largest web firms, is developing its own chips to be used in servers and autonomous cars under the brand name Kunlun.

Alibaba's semiconductor design division called T-Head has developed an AI inference chip called Hanguang 800. Inference is the method that follows the training of AI models because it refers back to the actual application of AI in the true world, equivalent to a chatbot responding to user queries.

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

Like Nvidia, Biren Technology develops a general-purpose GPU and has a software development platform for constructing applications on the hardware.

These chips are a part of Biren's Bili product line, designed to be used in data centers for AI training.

Last 12 months, Biren was placed on a U.S. blacklist often known as the Entity List, which restricts the corporate's access to certain American technologies.

Cambricon Technologies develops various forms of semiconductors, from those for training AI models to people who can run AI applications on devices reasonably than in data centers.

However, the corporate continues to record significant losses and According to reports laid off employees last 12 months, the South China Morning Post reported.

Cambricon Technologies can be on the US Entity List.

Moore Threads, founded in 2020, develops GPUS Designed to coach large AI models.

MTT KUAE is the corporate's data center product that houses its GPUs. The company's mission is to change into a “global GPU leader,” it said in a press release on its website.

Big brands are also supporting the corporate. TikTok owner ByteDance is an investor alongside major enterprise capital firms equivalent to Sequoia and GGV Capital.

Moore Threads can be on the US Entity List.

Enflame Technology is one other startup in China that goals to position itself as a domestic alternative to Nvidia. The company develops chips for data centers that concentrate on AI training and processing.

Tencent, considered one of China's largest technology firms, is an investor in Enflame.