Chinese company DeepSec is developing its own artificial intelligence chip, Reuters reported, citing informed sources. This move reflects a growing trend among Chinese AI companies to reduce their reliance on foreign chips, particularly given the ongoing US restrictions on exporting advanced semiconductors to China.

The initiative is still in its early stages, the report stated, with the company working with external partners and employing engineers specializing in chip design. The anticipated chip will focus on reasoning—running models and responding to users after training—rather than necessarily training large models from scratch.

From models to hardware,

DeepSec has emerged as one of China's leading AI companies, having developed models that garnered global attention for their efficiency and ability to compete with larger models at relatively lower operating costs. However, the AI ​​story is no longer solely about the model itself, but also the infrastructure that powers it.

As the number of users and applications increases, the cost of inference, power consumption, and response time become critical factors. Therefore, many companies strive to improve the relationship between software and chips, designing models to leverage hardware characteristics or developing the hardware itself to meet the model's specific needs.

For DeepSec, owning a custom chip could grant it greater control over performance and cost, rather than relying entirely on off-the-shelf chips from Nvidia or Huawei. However, this also presents complex engineering and manufacturing challenges. This

move

comes within the broader context of technological tensions between Washington and Beijing. In recent years, the United States has imposed restrictions on the export of advanced AI chips to China, prompting Chinese companies to seek domestic alternatives or develop customized solutions.

DeepSec had previously relied on Nvidia H800 chips, originally designed for the Chinese market before the tightening of US restrictions. Reuters also reported earlier that the DeepSec V4 model was modified to run efficiently on Huawei's Ascend chips, indicating a growing reliance on domestic hardware within China's AI sector.

Developing a custom chip doesn't mean DeepSec will immediately abandon its current suppliers. Designing an AI chip is not enough; the company needs to manufacture it, provide advanced memory, and build a software system capable of running the models efficiently.