Kunlunxin M300 to Launch Next Year, Tianshi 1024-Card Super Node to Hit the Market Simultaneously
Kunlunxin M100 has been launched on schedule, and the flagship M300 will be launched next year, with upgrades in computing power, storage, interconnectivity and other dimensions; in 2028, multiple chips based on the all-new N architecture will also be launched. Baidu's Tianshi 1024-card super node is also planned to be brought to market next year; a single unit can support training of models with 5 trillion parameters, and can build a supercluster of up to 500,000 cards. Li Yanhong stated that inference will become the main driving force for the long-term sustained growth of AI chips.
Body · AI translation
Kunlunxin M300 to Launch Next Year Source: Semiconductor Industry Observer Published: 2026-11-12 17:38 Kunlunxin chip manufacturing Generate poster Kunlunxin M300 achieves capability upgrades in dimensions such as computing power, storage, and interconnect.
Just now, Kunlunxin announced its next product plan. The M100 has already been launched on schedule, and the M300 will be launched next year. Among them, the M100 focuses on extreme cost-effectiveness; while the flagship M300 upgrades capabilities in multiple dimensions such as computing power, storage, and interconnect. By 2028, multiple chips based on the all-new N architecture will also be brought to market, matching differentiated application scenarios.
Advancing in parallel with the chips is the expansion of cluster scale. Baidu Group Executive Vice President Shen Dou revealed that the Tianchi 1024-card super node is also planned to be launched next year. A single set can support training of models with 5 trillion parameters and has the ability to scale to 10,000 cards; through the upgrade of a new-generation network architecture, a supercluster of up to 500,000 cards can be built to support next-generation ultra-large-scale model training.
Baidu founder Robin Li, when talking about AI chips, said that in the past, AI computing demand was mainly driven by training, whereas now it is a landscape where training and inference are equally important. In particular, the scale-up of agent applications has driven an explosive surge in inference computing demand. "We believe that inference will become the main driving force for the long-term sustained growth of AI chips."
On the product side, Kunlunxin has successively launched multiple generations of AI chip products. The first-generation Kunlunxin AI chip is designed based on Kunlunxin's self-developed XPU architecture, targets cloud inference scenarios, supports general AI algorithms, and delivers efficient and stable performance metrics in algorithms for computer vision, speech recognition, natural language processing, and recommendation. At present, this chip has been deployed in the tens of thousands in businesses such as Baidu Search and Xiaodu, empowering fields including the internet, industrial manufacturing, smart finance, and smart transportation.
The second-generation Kunlunxin AI chip is designed based on the new-generation self-developed Kunlunxin XPU-R architecture and focuses on generality and ease of use. Compared with first-generation products, the general-purpose computing core computing power of the second-generation Kunlunxin AI chip has increased by 2-3 times, providing strong AI computing power for data centers.
In 2024, the third-generation Kunlunxin AI chip entered mass production, meeting the higher requirements of AI systems in the large-model era. In February 2025, Baidu AI Cloud successfully lit up the Kunlunxin third-generation P800 10,000-card cluster; in April of the same year, Baidu announced that the 30,000-card cluster of Kunlunxin P800 had been successfully lit up. Up to now, Kunlunxin has delivered multiple 10,000-card-level P800 clusters, implemented in industries such as the internet, finance, energy, and manufacturing, with delivery scale ranging from dozens of cards to tens of thousands of cards, ranking among the top in domestic AI chip market share.
In 2025, Kunlunxin revealed a new generation of products, including two products: the Kunlunxin M100, optimized for large-scale inference scenarios, and the Kunlunxin M300, aimed at ultra-large-scale multimodal model training and inference tasks. In September this year, the China Information Security Evaluation Center and the State Secrecy Science and Technology Evaluation Center issued the "Security and Reliability Evaluation Results Announcement (2026 No. 3)". The security and reliability level of the Kunlunxin M100 is Class I, and the result is valid for three years from the date of publication.
All three generations of Kunlunxin chips are based on the self-developed Kunlunxin XPU architecture released in 2017. This architecture is the core chip architecture developed by the Kunlunxin technology team for general AI computing. Starting from the actual needs of AI implementation, it iterates according to the requirements of complex and cutting-edge artificial intelligence scenarios, and is committed to providing developers with a general, easy-to-use, and high-performance source of computing power.
According to IDC data, among the leading domestic AI chip manufacturers in 2024, Huawei Ascend ranked first with 640,000 chips, followed by Kunlunxin, with annual shipments reaching 69,000 chips. In terms of market share, Kunlunxin accounted for about 2.5% of global shipments and more than 8% of China's domestic market share. In 2025, Huawei Ascend ranked first with shipments of 812,000 chips, T-Head ranked second with 265,000 chips, and Kunlunxin and Cambricon each shipped 116,000 chips, tied behind them.
On the capital side, Kunlunxin has successively advanced a Hong Kong stock listing application and STAR Market listing tutoring. On January 1, Kunlunxin, through joint sponsors, submitted an A1 listing application form to the Hong Kong Stock Exchange in confidential form, intending to spin off and list on the main board of the Hong Kong Stock Exchange. In May 2026, Kunlunxin also completed the filing for STAR Market IPO tutoring, with CICC serving as the tutoring institution. The filing materials show that Baidu (China) Co., Ltd. holds 57.67% of Kunlunxin's shares and is its controlling shareholder.
Earlier this month, there were rumors that Kunlunxin's listing time might be delayed. Baidu responded that the relevant rumors and reports are untrue. At present, Kunlunxin is still in the confidential filing stage. In accordance with regulatory requirements, major non-public information such as the listing process cannot be disclosed for the time being.
This article is reproduced from media reports or online platforms and represents the author's personal stance or views. Our reproduction is solely for sharing and does not mean that we approve of or endorse it. If the source attribution is incorrect or your lawful rights and interests are infringed, please contact customer service in a timely manner. As a neutral platform service provider, we will promptly correct, delete, or handle it in accordance with the law. Comments No user comments yet Latest News More Related Recommendations
Beyond Bandwidth, AI Inference Is Reshaping the Memory Wall 2026-11-12 17:53
Automakers' Self-Developed Chips Begin External Supply, Can They Open New Markets? 2026-11-12 17:48
Bank of America: TSMC's 2027 Capital Expenditure Expected to Reach US$90 Billion, Advanced Process Expansion Accelerates 2026-11-12 17:46