DeepSeek and Zhipu AI Lead Chinese Custom Silicon Push Amid Global Inference Chip Wave

DeepSeek and Zhipu AI Lead Chinese Custom Silicon Push Amid Global Inference Chip Wave

The global race for custom AI silicon has intensified as leading Chinese AI labs seek to bypass hardware bottlenecks, lower serving costs, and reduce their reliance on Nvidia and Huawei. On July 7, 2026, a Reuters exclusive revealed that Chinese AI champion DeepSeek is developing its own custom AI chip designed specifically for model inference.

DeepSeek's Custom Inference Silicon Program

DeepSeek's custom chip program began in mid-2025 and is tailored for the inference stage—running trained models to generate user responses—rather than the intensive training phase. This aligns with global trends where AI developers seek to optimize hardware specifically for the massive compute demands of running reasoning models (such as DeepSeek-R1 and V4) at scale.

To accelerate the program, DeepSeek has quietly increased its private recruitment of chip-design engineers and has initiated discussions with foundry, chip-design, and high-bandwidth memory (HBM) companies.

"Chinese startup DeepSeek is developing its own AI chip, according to three people familiar with the matter, a push that could reduce its reliance on Nvidia and Huawei chips, which it has depended on to train and run its globally popular models... The chip is designed for inference — the stage of AI computing in which a trained model generates responses for users — rather than for training new models, the sources said."

Geopolitical and Supply Chain Hurdles

DeepSeek’s silicon ambitions face steep hurdles due to stringent U.S. export controls. While U.S. firms like OpenAI—which recently unveiled its custom Jalapeño inference chip built with Broadcom—have access to leading-edge manufacturing and high-bandwidth memory, Chinese firms are legally blocked from advanced foundries (such as TSMC's leading nodes) and advanced memory components.

Consequently, DeepSeek has had to lean heavily on domestic hardware. In April 2026, DeepSeek adapted its V4 model to run entirely on Huawei's Ascend chips. However, developing in-house custom silicon represents a strategic bid to secure hardware independence as domestic chip supply remains constrained¹.

"Designing a competitive AI chip typically takes years and significant capital. Manufacturing poses another hurdle as the U.S. bans Chinese designers from accessing the most advanced overseas foundries, while separate U.S. curbs have cut China's access to high-bandwidth memory, a component critical to AI inference chips."