[News] Baidu Rolls Out Kunlun Roadmap, M100 & M300 AI Chips Arrive 2026–2027

As China pushes to catch up with the U.S. in the AI race, local tech giants are answering the call with a steady stream of AI chip launches. Huawei has unveiled its Ascend AI chips, planning regular releases through 2028, while Baidu, at its World Conference on November 13, revealed it will roll out new products every year for the next five years, according to China.com.

IT Home highlights that Baidu’s new Kunlun generation includes two chips: the M100, optimized for large-scale inference, launching in 2026, and the M300, designed for training and inference of massive multimodal models, arriving in 2027. The report adds that the Kunlun N-series chips are slated for a 2029 launch.

Inside Baidu’s Massive Tianchi AI Supernodes

In terms of the supernode—a massive AI computing unit—the reports note that Baidu’s Tianchi 256 and 512 are set to launch next year. IT Home points out that Tianchi supernodes at the thousand-card scale are set to launch in 2028, adding that Baidu plans to power its Baige platform with a one-million-card Kunlun single-cluster deployment in 2030.

Notably, Wallstreetcn points out that at the event today, Baidu’s Kunlun supernode, which has already been deployed at large scale within the company, made its first public appearance. This in-house developed technology can boost single-card performance by 95% and increase single-instance inference performance by up to 8x, the report notes.

Wallstreetcn suggests that the Baidu Tianchi 256 supernode could support ultra-fast interconnection of up to 256 cards. The solution’s inter-card interconnect total bandwidth is four times higher than the supernode Baidu released in April this year, with performance improvements exceeding 50%, as noted by the report. For mainstream model inference tasks, single-card token throughput increases by 3.5x, according to Wallstreetcn.

According to Chinese media outlet C114, leveraging Kunlun chips and the Baige AI computing platform, Baidu Intelligent Cloud provides enterprises with high-performance, scalable AI computing power, and has ranked first in China’s AI cloud market for six consecutive years.

It is worth noting that prior to Baidu, Huawei revealed a multi-year roadmap for its Ascend AI chips in September. According to MyDrivers, the rollout kicks off with the Ascend 950PR in Q1 2026, followed by the 950DT in Q4 2026, the Ascend 960 in Q4 2027, and the Ascend 970 in Q4 2028. Notably, the report highlighted that the 950 series will feature Huawei’s in-house HBM, a key differentiator for performance.

Read more

• [News] Huawei Unveils Ascend 950 with In-House HBM in 2026, Touts SuperPoD to Rival NVIDIA

(Photo credit: Baidu’s X)