Baidu Introduces New AI Processors To Advance China’s Semiconductor Independence

Key takeaways

Chinese technology giant Baidu announced two new artificial intelligence semiconductors and supercomputing products at its annual Baidu World conference on Thursday, positioning itself as a key player in China's push for technological self-sufficiency in the face of escalating U.S. trade restrictions.

The Beijing-based company introduced the M100 and M300 chips, designed by its semiconductor unit Kunlunxin Technology, as solutions to provide Chinese enterprises with powerful, cost-effective computing alternatives to restricted American processors.

New chips target different AI workloads

The M100, scheduled for release in early 2026, focuses on inference operations, the process where AI models make predictions and respond to user requests.

The chip is specifically engineered to enhance efficiency in models using the mixture-of-experts technique, a method that activates only relevant portions of a neural network for specific tasks.

The M300, slated for early 2027, offers both training and inference capabilities.

Training involves building AI models by analyzing patterns in massive datasets, while inference applies those trained models to real-world applications. According to Baidu, the M300 is tailored for developing super-large multimodal models with trillions of parameters.

Shen Dou, executive vice-president at Baidu and president of its cloud unit, stated the chips would provide "powerful, low-cost and controllable AI computing power" to support China's AI self-sufficiency efforts, according to the South China Morning Post.

Supernode systems to compensate for chip limitations

Beyond individual processors, Baidu unveiled two supernode products that leverage advanced networking to link multiple chips together.

These systems aim to overcome the performance limitations of individual processors by creating large-scale computing clusters.

The Tianchi 256, incorporating 256 of Baidu's P800 chips, will become available in the first half of 2026.

An upgraded Tianchi 512 version, using 512 chips, is planned for the second half of 2026. Baidu claims the Tianchi 256 will deliver more than 50 percent performance improvement compared to its previous cluster architecture.

This supernode approach mirrors strategies employed by Huawei, which has deployed its CloudMatrix 384 system comprising 384 Ascend 910C chips.

Industry observers consider Huawei's solution more powerful than Nvidia's GB200 NVL72, one of the American chipmaker's most advanced system-level products.

Rising competition in China's domestic chip market

The announcements come as U.S.-China tensions have intensified, with restrictions on exports of advanced American AI chips to Chinese companies, accelerating efforts by Chinese firms to develop domestic alternatives.

Baidu has been investing in proprietary chip development since 2011, giving it a head start in the race for semiconductor independence.

The company faces competition from multiple domestic players.

Huawei's HiSilicon division continues advancing its Ascend AI chip series, while startups including Cambricon Technologies, MetaX Integrated Circuits, and Biren Technology are developing graphics processing units for AI training applications.

Baidu's stock gained 0.8 percent in Hong Kong trading Thursday morning following the announcements.

The company also unveiled an updated version of its Ernie large language model, claiming the new iteration excels not only in text processing but also in analyzing images and videos.

Read more: