Chinese search engine company Baidu has activated a 30,000-chip cluster of its own P800 Kunlun semiconductors, which the company claims can support the training of artificial intelligence (AI) models at the level of DeepSeek. The development is one of a series of advances by Chinese companies as they seek to develop an AI-based infrastructure that doesn’t require U.S. components.

Robin Li, CEO of Baidu, announcing the development at the company’s developer conference, said the cluster of Kunlun chips can enable the training of AI models with hundreds of billions of parameters. Alternatively, the cluster could support a thousand companies simultaneously fine-tuning models with billions of parameters, Li said.

Baidu unveiled this third generation Kunlun chip in February 2025, and it is considered China’s first domestically-developed large-scale AI computing deployment. The Kunlun is currently in use by Chinese internet companies and banks, and Baidu touts the chip as an AI processor for diversified workloads. It’s used in AI models that power NLP, HPC, vision and speech, and autonomous driving use cases. Customers can access the Kunlun chip cluster through the Baidu cloud platform.

Among the major Chinese tech companies, Baidu has been a leader in AI investment, even among the many Chinese companies that poured investment into AI after OpenAI launched ChatGPT in 2022. As part of its AI efforts, the company debuted a ChatGPT-style chatbot, dubbed Ernie, though the bot has gained only modest adoption in comparison with U.S.-based AI bots.

 Baidu’s activation of the 30,000-chip Kunlun cluster is a major milestone in a process that began in 2023, when the company placed its first chip order from Huawei instead of NVIDIA, a shift the company made in response to tighter restrictions on chip sales to China by the Biden administration. Although the order, valued at 450 million yuan ($61.83 million), represented only a minimal amount compared with the thousands of chips that Chinese companies were ordering from NVIDIA, this small action reflected a growing trend of Chinese companies developing non-U.S. sources of AI infrastructure.

That Baidu is using chips of its own design is significant not just technologically, but it also has ramifications for the trade war between U.S. and China. China, that is less reliant on U.S. resources, has less need to compromise on tariff negotiations.

However, the supply chain for the Kunlun chip appears to be facing a challenge. Kunlun is fabricated by Samsung, based in South Korea. In January of this year, Samsung announced that it had halted its supply of AI chips to Baidu due to U.S. export restrictions aimed at China.

Since that announcement, neither Samsung nor Baidu has provided more information, so presumably the chips in the current Kunlun cluster were purchased before Samsung ceased sales. Now it appears that Samsung, with an uncertain tariff situation going forward, faces uncertainty in its partnership with Baidu.

In any case, the Kunlun chip, while it represents a major advance for the Chinese chip sector, does not offer performance comparable with the newest releases of NVIDIA semiconductors. The previous Kunlan chip, the Kunlan II, was said to offer performance roughly approximate with NVIDIA’s A100.

Even if the newest Kunlun is a major leap forward from its predecessor, it does not support AI processing at the level of the most recent NVIDIA Blackwell chips.

Still, the Chinese efforts at homegrown chips are gaining on their U.S. competitors year by year, and as DeepSeek’s use of less capable chips demonstrated, advanced AI models don’t always require the latest chips.