Baidu’s AI Chip Arm Kunlunxin Moves Toward Hong Kong IPO as China Accelerates Semiconductor Self-Reliance
Baidu has confidentially filed for a potential Hong Kong listing of Kunlunxin, its artificial intelligence chip subsidiary, signaling growing confidence in China’s domestic AI semiconductor sector. While details of the offering remain undecided and regulatory approvals are still required, the move reflects Baidu’s intention to unlock Kunlunxin’s standalone value while retaining majority ownership.
The planned spin-off comes as Beijing accelerates efforts to build a self-sufficient AI and semiconductor ecosystem amid tightening U.S. export controls on advanced chips from companies like Nvidia. In response, Chinese chipmakers are increasingly turning to capital markets for funding, with several peers announcing IPO plans in recent months.
Founded in 2012, Kunlunxin sits at the core of Baidu’s “full-stack” AI strategy, supporting data centers, cloud computing, and deployment of the company’s Ernie AI models. While Baidu still relies on Nvidia for high-end training, Kunlunxin’s chips are increasingly used for inference and production workloads, where cost efficiency and supply stability matter most.
Kunlunxin has also expanded beyond internal use, growing sales to third-party customers including state-backed cloud and telecom players. Analysts view its strong software compatibility as a key advantage, enabling easier migration from Nvidia-based systems. Reuters previously reported that Kunlunxin’s revenue exceeded 3.5 billion yuan last year and reached break-even, with external customers expected to contribute a majority of sales going forward.
Despite rising traction, Kunlunxin is not seen as a full substitute for Nvidia’s most advanced processors, given China’s ongoing constraints in leading-edge chip manufacturing. Instead, it is positioned as part of a broader domestic ecosystem alongside players such as Huawei, Cambricon, and Alibaba, supporting Beijing’s push to reduce reliance on foreign AI hardware.











