Two of China's leading artificial intelligence developers, DeepSeek and Zhipu AI, are reportedly building their own custom chips rather than relying entirely on Nvidia's graphics processing units, according to reports from Pandaily and HackerNoon.
The move is part of a broader push by Chinese AI model developers to bring chip design in-house. According to Dataquest India, Zhipu AI is exploring ASICs — application-specific integrated circuits, or chips designed for particular AI workloads — as an alternative to Nvidia GPUs, and has approached several Chinese chip designers about the effort.
DeepSeek's work is reportedly focused on inference chips. Inference is the stage where a trained AI model is actually put to use answering questions or generating text, as opposed to the more compute-heavy training phase. According to HackerNoon, DeepSeek's inference chips would push AI computing power further into the deployment stack — the layer where models run in real-world applications.
Custom silicon designed for a specific task can be cheaper and more power-efficient at scale than general-purpose GPUs, though it typically requires large upfront investment and access to advanced chip manufacturing.
The sources here are early reports, and the companies have not published detailed specifications or timelines. What is described is exploration and development rather than finished, shipping products.
Why it matters: Nvidia's GPUs have been the default engine of the global AI boom, and U.S. export limits have constrained Chinese firms' access to the most advanced ones — so if DeepSeek and Zhipu can design capable chips of their own, it signals China's AI industry is working to reduce its dependence on a single dominant foreign supplier.