Role in Baidu’s AI Strategy
Kunlunxin, founded in 2012, designs application-specific chips for data centers, cloud services and edge computing. The unit underpins Baidu’s ambition to operate a “full stack” of AI capabilities, spanning hardware, servers, data centers, large language models and consumer applications such as the Ernie chatbot.
Although Baidu remains a significant purchaser of high-end graphics processors from U.S. supplier Nvidia, the company has been deploying Kunlunxin products alongside Nvidia components in its own data centers. The in-house chips are used to run and scale Baidu’s proprietary AI models, reducing exposure to external supply constraints.
Over the past two years Kunlunxin has moved beyond servicing only its parent company, selling processors to telecom carriers, cloud providers and other corporate clients. Industry analysts characterize the chips as among the most widely adopted domestic AI accelerators, noting their compatibility with mainstream development frameworks and the relative ease of shifting workloads from Nvidia hardware.
Financial Performance and Market Interest
Reuters previously reported that Kunlunxin generated more than 3.5 billion yuan (about US$500 million) in revenue last year and broke even on a net basis. External customers are projected to account for a majority of its sales by 2025. The company has already secured chip orders exceeding 1 billion yuan from suppliers to China Mobile, one of the country’s largest wireless operators.
Kunlunxin’s most recent funding round, which included an investment by China Mobile, raised over 2 billion yuan and valued the business at approximately 21 billion yuan. Analysts at JPMorgan have forecast that annual chip sales could rise sixfold to 8 billion yuan by 2026, citing growing demand from government, telecom and state-owned cloud clients.
According to Hong Kong Exchanges and Clearing, the bourse has seen a pick-up in filings from mainland semiconductor designers as companies seek to capitalize on investor appetite for hardware tied to generative AI. Moore Threads and Biren Technology have both announced initial public offering plans in recent months.
Geopolitical and Policy Backdrop
Baidu’s move comes amid heightened technology competition between the United States and China. Washington has tightened export controls on cutting-edge AI chips produced by Nvidia and other American companies, while Beijing has urged domestic buyers to prioritize local alternatives and has directed billions of dollars in public funding toward chip research and manufacturing.
Industry specialists view Kunlunxin as a complementary rather than complete substitute for top-tier foreign processors. The company’s chips are considered particularly effective for AI inference workloads, where stable supply and cost efficiency are often prioritized over maximum performance. Persistent limitations in China’s most advanced fabrication processes mean Kunlunxin cannot yet fully match Nvidia’s leadership in training complex AI models.
Nonetheless, the unit is expected to play a central role in the broader domestic ecosystem being built around AI computing. Alongside platforms from Huawei’s Ascend division, Cambricon Technologies and Alibaba Group, Kunlunxin is part of a multivendor strategy designed to lower national dependence on overseas semiconductors.
Next Steps
Details of Kunlunxin’s share offering—including the number of shares to be issued, the price range and potential cornerstone investors—are likely to emerge once Hong Kong regulators complete their review. If approved, the deal would provide Baidu with a new source of capital to fund research, expand production capacity and compete in the increasingly crowded AI hardware market.
Baidu did not disclose an expected listing date, but market participants anticipate that the company will monitor conditions in the Hong Kong equity market and broader semiconductor sector before finalizing terms. Until regulatory clearance is obtained, Kunlunxin will continue to operate under Baidu’s corporate structure and supply both internal and external customers.
Crédito da imagem: NurPhoto via Getty Images
