Chinese technology giants Alibaba and Baidu have begun using their own chips to train artificial intelligence models, a move that shows Beijing’s drive for self-sufficiency in critical technologies and poses a growing challenge to Nvidia’s dominance in the Chinese market.
Alibaba and Baidu are reportedly ditching Nvidia’s processors and utilizing home-made solutions, with Alibaba training smaller AI models with its in-house processors. Baidu is also testing the waters with its latest Kunlun P800 chip on more advanced versions of its Ernie model.
However, it highlights the progress and seriousness that Chinese firms are putting into making chip designs and the urgency created by U.S. export restrictions that have curbed access to Nvidia’s most advanced processors.
Breaking reliance on U.S. hardware
Over the years, Chinese companies building large-scale AI systems have relied heavily on Nvidia’s hardware, particularly the H100 and its successors.
Nvidia is the market leader in the GPU space, known to be part of the critical infrastructure components in the AI boom. However, Chinese companies can’t get their hands on the most advanced Nvidia chips, as the U.S. government has progressively tightened the noose around getting those chips, citing national security reasons.
However, last month, Cryptopolitan reported that Jensen Huang, CEO of Nvidia, said he was discussing with Washington to push for the sale of some advanced versions of Nvidia’s next-generation Blackwell chips. Despite this, it still seems that the most powerful models remain off limits.
The restrictions have accelerated Beijing’s push to develop a domestic semiconductor ecosystem, backed by billions of dollars in state subsidies and pressure on national champions to cut reliance on U.S. suppliers.
Alibaba’s Zhenwu and Baidu’s Kunlun chips
Alibaba’s semiconductor arm has designed the Zhenwu processing unit to support cloud-based AI workloads.
Baidu has been working with its own Kunlun line of AI accelerators. The latest P800 version is being used to train newer versions of Ernie, the company’s flagship large language model. Baidu has also positioned Kunlun as both a cost-saving tool and a hedge against foreign sanctions.
For now the switch is partial as both companies still continue to rely on Nvidia chips for their most advanced systems, especially those involving very large training runs where performance gaps remain. But the willingness work with Chinese-made alternatives in production environments is a major step in the move away from American tech.
Implications for global AI competition
For Nvidia, the development is a warning sign as China has been its largest overseas market. A successful move away from the America chipmaker by Chinese firms to home-grown GPUs could end demand for U.S. chips and greatly impact Nvidia’s market share and balance book, even when import restrictions are removed.
For Alibaba and Baidu, the ability to design and control their own hardware could lower costs and help them align with Beijing’s industrial policies.
Matching Nvidia is still a longshot, but it’s looking closer with the latest development. The firms still have to match Nvidia’s performance and ecosystem support, which boasts advanced software frameworks such as CUDA.
Get up to $30,050 in trading rewards when you join Bybit today
