Baidu has unveiled two new artificial intelligence (AI) chips and a set of advanced computing products, marking another strategic push in China’s bid for tech sovereignty.
The announcement, made at the Baidu World 2025 technology conference, comes amid intensifying US-China chip tensions and rising restrictions on exports of advanced American AI semiconductors to Chinese companies.
The move reflects a growing national trend: accelerating the domestic development of high-performance processors to reduce reliance on foreign technologies.
The AI chip space in China is no longer just about innovation. It is becoming a critical component of the country’s economic resilience and digital infrastructure.
Domestic chip design enters a new phase
The two new chips, M100 and M300, are designed for AI workloads. Baidu’s M100, an inference-focused chip, is set for release in early 2026.
Inference, which uses trained AI models to generate real-time outputs, is central to applications such as recommendation systems, chatbots, and intelligent search.
Meanwhile, the M300 will support both training and inference. Training enables AI models to learn from large volumes of data, making this chip a versatile option.
Baidu has scheduled its launch for early 2027. These products reinforce the company’s commitment to long-term semiconductor development, which began in 2011.
Baidu is not alone in this race. Rival Chinese tech firm Huawei has already drawn industry attention with its CloudMatrix 384, a supercomputing cluster that uses 384 of its Ascend 910C chips.
The system is viewed by some observers as more powerful than Nvidia’s GB200 NVL72, one of the most advanced offerings currently available from the US-based chipmaker.
Supernodes spotlight China’s infrastructure build-out
In addition to its new chips, Baidu introduced two high-capacity computing platforms known as supernodes.
These systems aim to boost overall performance by networking multiple chips together, addressing the limitations of standalone units.
The first product, Tianchi 256, will feature 256 of Baidu’s P800 chips and is expected in the first half of 2026.
A larger version using 512 chips will follow in the second half.
The development of such systems shows a clear focus on scalable infrastructure that aligns with national AI priorities.
Supernodes offer not only high computational throughput but also represent a shift in how Chinese companies are approaching hardware design. The focus is now on modular, scalable systems that can compete globally.
Ernie model upgraded for vision, video, and text
Another key highlight from the event was the latest version of Baidu’s large language model, Ernie.
The company said the new version is capable of handling not only text inputs but also image and video-based tasks.
This positions it more directly against multimodal models developed by international competitors and fits within China’s broader ambitions to advance generative AI.
Multimodal models are essential for next-gen applications, from video content moderation to AI-powered digital assistants. Baidu’s emphasis on integrating these capabilities signals a strategy that moves beyond catching up.
It marks the start of a new phase where Chinese firms are building products that challenge international benchmarks.
The post China ramps up AI autonomy as Baidu reveals new chips and supernodes appeared first on Invezz
