Baidu Targets a Hong Kong IPO for AI Chip Unit Kunlunxin as China Races to Replace Nvidia

Baidu confirmed Kunlunxin filed confidentially for a Hong Kong listing on January 1, setting up a potential spin-off IPO that could fund one of China’s most commercially active AI chip efforts. The move lands amid tighter U.S. chip restrictions, Beijing’s push for semiconductor self-sufficiency, and a strong return of capital to Hong Kong markets.

Key Points

• IPO filing: what happened and what is unknown

  Kunlunxin submitted a confidential listing application to the Hong Kong Stock Exchange on January 1, 2026, confirmed by Baidu on January 2. IPO size and structure are not finalized, approvals are required, and Baidu says the spin-off may not proceed. Baidu keeps a controlling stake of about 59% and expects Kunlunxin to remain a subsidiary after listing.

• Why now: policy pressure + funding demand

  The listing plan fits China’s push to build domestic AI compute supply as U.S. export controls constrain access to advanced chips. State-backed capital and procurement policy have increased demand for local chips and encouraged more public listings among semiconductor and AI infrastructure firms.

• Why Kunlunxin matters inside Baidu

  Founded in 2012, Kunlunxin supports Baidu’s “full stack” AI strategy across chips, servers, data centers, models, and applications. Baidu still uses Nvidia heavily for top-end workloads, but Kunlunxin chips are increasingly used in Baidu data centers running Ernie AI models, especially for inference and cost-sensitive deployment.

• Commercial traction: external sales becoming central

  Kunlunxin has become independently operated while remaining majority-owned by Baidu. External sales expanded over the past two years, with revenue projected above 3.5 billion yuan (≈$500 million) and a break-even trajectory. Third-party customers are expected to contribute more than 50% of revenue in 2025, shifting Kunlunxin from internal supplier to market-facing vendor.

• Proof points: contracts + valuation signals

  In 2025, Kunlunxin won orders exceeding 1 billion yuan tied to suppliers serving China Mobile. China Mobile also joined a funding round that raised 2+ billion yuan and valued Kunlunxin at about 21 billion yuan (≈$3 billion), a key reference point for investor expectations ahead of any IPO.

• Positioning versus Nvidia: where it competes and where it does not

  Analysts highlight Kunlunxin’s practical adoption and software compatibility with mainstream AI frameworks, easing workload migration from Nvidia stacks. Manufacturing constraints mean it is not a full replacement for Nvidia in frontier training, reinforcing its near-term edge in inference-heavy use cases for government, telecom, and state-linked cloud buyers prioritizing stable supply and cost.

• Market tailwind: Hong Kong IPO rebound

  Hong Kong raised $36.5 billion across 114 listings in 2025, more than triple $11.3 billion in 2024 (LSEG). China’s semiconductor financing also stayed hot: 20 semiconductor IPOs raised 45+ billion yuan last year, and the broader chip industry drew 185+ billion yuan across 1,419 financing cases, creating a receptive backdrop for a Kunlunxin float.

• Forward look: growth targets and ecosystem logic

  JPMorgan forecasts Kunlunxin chip sales could rise sixfold to 8 billion yuan by 2026. Strategically, Beijing appears to be building a multi-vendor domestic AI compute ecosystem, Kunlunxin alongside Huawei Ascend, Cambricon, Alibaba, and others, rather than betting on a single national champion.

Why This Matters

This IPO plan is a live indicator of how AI infrastructure is being reshaped by geopolitics and capital markets. If Kunlunxin lists, it could accelerate domestic AI compute capacity in China, widen the investor base for Chinese chipmakers, and intensify competition in inference-focused data center deployments, especially where Nvidia supply is constrained.

This article was drafted with the assistance of generative AI. All facts and details were reviewed and confirmed by an editor prior to publication.