Author｜Chen An

Editor｜Chen Xiaoran

On May 7, an announcement regarding coaching filings on the website of the Beijing Securities Regulatory Bureau once again thrust Baidu's Kunlunxin into the limelight.

A+H Synchronous Parallel Approach

This AI chip unicorn, which originated from Baidu's internal chip team, officially commenced IPO coaching for the STAR Market just four months after submitting a confidential application for a main board listing on the Hong Kong Stock Exchange (HKEX) in January 2026. China International Capital Corporation (CICC) serves as the sole coaching institution, with Han Kun Law Offices and Ernst & Young Hua Ming LLP handling legal and auditing services, respectively.

The progress of Kunlunxin's HKEX listing has not been hindered by the initiation of STAR Market coaching and continues to advance as originally scheduled. This indicates that Kunlunxin will adopt the industry's rare 'A+H synchronous parallel' capital route, rather than the traditional 'A before H' or 'H before A' models.

The simultaneous establishment of dual-market access aims to capitalize on the STAR Market's 'hard technology' attributes to secure valuation premiums and policy benefits, while also attracting global long-term capital through HKEX's international platform to ensure sustained financing capabilities.

At the equity structure level, Baidu has consistently maintained firm control, currently holding a 57.67% direct stake in Kunlunxin. Baidu's founder, chairman, and CEO, Robin Li, indirectly holds approximately 11% through various entities, including Beijing Xin Shiji and Tianjin Jiuzhang Kunlun.

After five years and six rounds of financing, Kunlunxin's shareholder base has expanded to 57 institutions, including national team capital such as the China Internet Investment Fund and the Beijing AI Industry Investment Fund, as well as industrial and financial capital from BYD, China Mobile Innovation (a subsidiary of China Mobile), and CITIC Securities Investment. The equity structure has become increasingly diversified.

Discrepancies in capital market valuations have become one of the focal points of this IPO. During its independent financing in 2021, the first-round valuation reached RMB13 billion, setting a record for initial financing in the domestic AI chip industry. After completing Series D financing in July 2025, the post-investment valuation soared to RMB21 billion.

Foreign investment banks offer even more varied valuation ranges. Goldman Sachs conservatively estimates the value of Baidu's stake in Kunlunxin at USD3-11 billion, while Macquarie provides an optimistic forecast of USD16.5 billion. Morgan Stanley even predicts a sixfold revenue surge to RMB8.3 billion in 2026.

Centralized Pricing Test

The significant tension between 'real-world performance valuation' and 'domestic substitution imagination valuation' makes Kunlunxin's IPO more than just a simple asset spin-off for Baidu; it represents a centralized pricing test for domestic AI chip assets in the capital market.

Emerging from Baidu's Intelligent Chip and Architecture Department, Kunlunxin has undergone 15 years of technological accumulation, completing a crucial transformation from an internal cost center to an independent commercial entity. Its revenue growth trajectory and product deployment capabilities form the core confidence supporting its capital market sprint.

Tracing its development, Kunlunxin's inception stemmed from Robin Li's cost considerations, believing that 'building is superior to buying.'

In 2011, faced with escalating computational demands from Baidu's search business and overseas chip prices reaching USD20,000 per unit, Baidu was compelled to initiate internal FPGA AI accelerator research and development. In 2018, it launched the first-generation Kunlun AI chip, achieving large-scale deployment in 2020. In April 2021, it formally began independent operations, with Baidu's chief chip architect, Ouyang Jian, appointed as CEO.

Today, Kunlunxin has established full-stack technological capabilities encompassing 'chips + boards + clusters + software stacks,' with over 400 invention patent applications. Its product matrix covers all scenarios, from cloud training and inference to edge computing.

Its current flagship product is the P800 training chip, which entered mass production in 2024. Utilizing a self-developed XPU-P architecture, it delivers 345 TFLOPS of FP16 computational power and 20%-50% higher memory bandwidth than comparable mainstream GPUs. It successfully powered China's first 30,000-card self-developed AI cluster, capable of simultaneously supporting multiple large models with hundreds of billions of parameters. It is deeply integrated into core Baidu businesses such as Baidu Search, Xiaodu, and ERNIE.

The product iteration roadmap is clear: In 2026, it will launch the M100 chip optimized for large-scale inference scenarios, followed by the M300 chip for ultra-large-scale multimodal models in 2027. The accompanying Tianchi 256 and Tianchi 512 supernodes will also roll out this year, with a single Tianchi 512 supernode capable of training trillion-parameter models.

In terms of commercialization, Kunlunxin has achieved breakthroughs from internal use to external marketization. In 2025, it won bids for China Mobile's AI server procurement project, securing 70%, 70%, and 100% shares in three inference-type equipment packages, earning a ten-billion-yuan order. This marks its entry into national-level computational infrastructure procurement systems.

Additionally, Kunlunxin has penetrated leading government, enterprise, and industrial clients such as China Merchants Bank, China Southern Power Grid, and BYD, continuously expanding its external customer base.

IDC data shows that in 2025, China's total AI accelerator card shipments reached 4 million units, with domestic vendors accounting for 41%. Kunlunxin ranked fifth overall and third among domestic vendors with 116,000 units shipped, tied with Cambricon, trailing only Huawei Ascend and Alibaba T-Head.

Over Half of Revenue Comes from Outside Baidu Ecosystem

Financially, Kunlunxin achieved approximately RMB2 billion in revenue in 2024 with a net loss of RMB200 million. In 2025, revenue is expected to surpass RMB3.5 billion, potentially achieving break-even, with over half of revenue coming from enterprises outside Baidu. Multiple securities firms predict 2026 revenue will climb to RMB6.5-8.3 billion, with external revenue accounting for an even larger share.

However, market skepticism about its independence persists. Early revenue and scenarios were highly tied to Baidu's ecosystem. While external customer expansion has shown breakthroughs, it remains in the cultivation phase. Whether it can continuously diversify its customer structure and establish self-sustaining capabilities independent of a single ecosystem will directly determine its valuation ceiling.

Kunlunxin's A+H sprint is not an isolated corporate capital move but a landmark event signaling the domestic AI chip industry's transition from policy-driven and ecosystem-driven growth to capital market validation. It carries the strategic value of China's computational autonomy substitution and the underlying logic of industry landscape restructuring.

Currently, the AI industry is experiencing explosive growth, with large models evolving from single-modal text to multimodal and world models, driving exponential growth in computational demand.

Frost & Sullivan predicts that China's AI chip market size will grow from RMB142.5 billion in 2024 to RMB1.3 trillion in 2029, with a compound annual growth rate of 54% from 2025-2029. GPU clusters with 100,000 or even 1 million cards are becoming the infrastructure entry tickets for next-generation large model competition.

Meanwhile, intensifying international supply chain uncertainties have transformed 'controllable computational power' from an industrial slogan into a procurement standard. Substitution demands in critical sectors such as telecommunications, finance, government, and industry are accelerating, presenting historic opportunities for leading enterprises like Kunlunxin.

Against this backdrop, domestic AI chip companies are entering a dense capitalization window. By the end of 2025, MetaX and Moore Threads listed on the STAR Market, soaring 693% and 425% on their debuts, respectively. Cambricon's market capitalization now exceeds RMB750 billion. Enflame Technology's STAR Market IPO has been accepted, while Biren Technology and Iluvatar CoreX opted for HKEX listings. Alibaba's T-Head has also circulated IPO rumors.

As the largest domestic AI chip company preparing to list, Kunlunxin's A+H listing will not only secure sufficient capital support for technological R&D and capacity expansion but also enhance brand influence and customer trust through capital markets, potentially securing a more favorable position in the industry's next consolidation phase.

However, from an industry perspective, Huawei Ascend remains the dominant leader, with domestic vendors each pursuing their strategies. Coupled with NVIDIA's price cuts to capture market share, the domestic camp may be the first to descend into a price war.

High valuations require concrete performance delivery. Revenue growth and customer structure optimization falling short of expectations will trigger valuation corrections.

The chip industry's characteristics of high investment and long cycles, combined with the stringent disclosure requirements and expectation management pressures of A+H listings, further amplify uncertainties.

Ultimately, Kunlunxin's IPO value lies not in benchmarking against NVIDIA but in whether it can externalize Baidu's ecosystem advantages into independent competitiveness.

The outcome of this capital examination will not only determine Kunlunxin's fate but also provide crucial references for the valuation system and development path of the domestic AI chip industry.