Baidu Unveils Ambitious AI Chip Roadmap, Targeting 1 Million-Card Cluster by 2030
00:00/00:00
Listen to this article
1x
Subscribe now
Listen to the full version
Listen to the full version
Baidu introduces plans to launch M100 and M300 AI chips at its annual conference in Beijing on Nov. 13, 2025.
Baidu Inc. on Thursday unveiled a sweeping five-year roadmap for its Kunlun AI chips, outlining new products through 2030 and major upgrades to its large-scale computing clusters.
Baidu said its M100 and M300 AI chips will debut in 2026 and 2027, respectively, with the M100 designed for inference and the M300 supporting both training and inference. The company also plans to launch its next-generation N-series in 2029 and reach a one-million-card Kunlun chip cluster by 2030.
You've accessed an article available only to subscribers
VIEW OPTIONS
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.
- RELATED
Share this article
Open WeChat and scan the QR code
- DIGEST HUB
Digest Hub
Back
Explore the story in 30 seconds
- Baidu announced a five-year AI chip roadmap, unveiling M100 (2026), M300 (2027), and next-gen N-series (2029), aiming for a one-million-card Kunlun cluster by 2030.
- Baidu introduced the 2.4-trillion-parameter Ernie 5.0 model, with performance comparable to top global competitors, and expanded supernode clusters, targeting 512-card nodes in 2026.
- Kunlun P800 clusters serve over 100 customers; Baidu leverages flexible overseas production and sustained Nvidia software compatibility amid ongoing AI chip export restrictions.
AI generated, for reference only
Who’s Who
- Baidu Inc.
- Baidu Inc. unveiled a five-year roadmap for its Kunlun AI chips, showcasing new products through 2030 and significant upgrades to its computing clusters. They announced M100 and M300 AI chips for 2026 and 2027, respectively. The company also introduced its Ernie 5.0 large model, claiming it matches top global systems like Gemini-2.5-Pro and GPT-5-High in language and perception capabilities.
- Huawei Technologies Co. Ltd.
- Huawei Technologies Co. Ltd. is mentioned as one of the companies, alongside Alibaba Group Holding Ltd. and Inspur Co. Ltd., that are increasingly adopting large-scale multi-card "supernodes" to enhance system-level performance. This approach is being utilized in response to Nvidia Inc.'s H20 export restrictions in China. Huawei Technologies Co. Ltd. has also shifted entirely to domestic supply chains after being placed on U.S. trade blacklists.
- Alibaba Group Holding Ltd.
- Alibaba Group Holding Ltd. is mentioned as one of the companies, alongside Huawei Technologies Co. Ltd. and Inspur Co. Ltd., that are increasingly adopting large-scale multi-card "supernodes" to boost system-level performance. This approach is being utilized due to restrictions on Nvidia Inc.'s H20 exports, and because Chinese AI chip designers cannot access sub-7-nanometer manufacturing overseas.
- Inspur Co. Ltd.
- Inspur Co. Ltd. (浪潮智能科技) is a Chinese company that, alongside Huawei Technologies Co. Ltd., Alibaba Group Holding Ltd., and others, is increasingly adopting large-scale multi-card "supernodes." This approach aims to boost system-level performance for AI chips, particularly in response to Nvidia Inc.'s export restrictions.
- Nvidia Inc.
- Nvidia Inc. is mentioned in the article as having export restrictions on its H20 chips, impacting Chinese AI chip designers. Baidu Inc.'s M100 and M300 AI chips are developing in 2026 and 2027 where Baidu also stated that Kunlun's compatibility with Nvidia's software ecosystem helps reduce development costs.
- Cambricon Technologies Corp. Ltd.
- Cambricon Technologies Corp. Ltd. is a Chinese AI chip designer that, unlike Baidu's Kunlun Chip, has shifted entirely to domestic supply chains. This move occurred after the company was placed on U.S. trade blacklists.
AI generated, for reference only
What Happened When
- 2011：
- Kunlun Chip was founded.
- 2021：
- Kunlun Chip began mass production of its second-generation 7-nm chip.
- February 2025：
- Baidu activated its first 10,000-card Kunlun P800 cluster.
- Earlier in 2025：
- Baidu Cloud rolled out 32-card and 64-card Kunlun supernodes.
- April 2025：
- Baidu expanded Kunlun P800 cluster to 30,000 cards; most inference workloads began running on this cluster.
- By November 12, 2025：
- A 5,000-card P800 cluster used for training a multimodal model scaled to more than 10,000 cards.
- November 12, 2025：
- Baidu unveiled a five-year roadmap for Kunlun AI chips and introduced Ernie 5.0 at Baidu World.
AI generated, for reference only
- PODCAST
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新