Chinese artificial intelligence startup DeepSeek has suggested that China is close to unveiling homegrown “next generation” chips designed to power advanced AI systems, even as it rolled out an update to one of its flagship large language models.

In a post on its official WeChat account on Thursday, the company said its newly released V3.1 model uses the “UE8M0 FP8” precision format, tailored for next-generation domestic chips expected to launch soon.

FP8, or 8-bit floating point, is a processing format that improves efficiency in training and inference for large-scale deep learning models.

DeepSeek’s remarks come as Beijing ramps up efforts to reduce reliance on United States technology amid Washington’s tightening restrictions on exports of advanced semiconductors.

Just two weeks ago, Chinese regulators reportedly urged local AI developers to prioritize domestic chip alternatives over Nvidia’s graphics processing units, the current industry standard.

Analysts note that while Chinese firms still lag behind Nvidia in scale and performance, companies such as Huawei have made strides in narrowing the gap.

DeepSeek, which made headlines earlier this year with its R1 reasoning model, has already proven capable of delivering performance on par with Western rivals despite limited access to Nvidia’s most advanced training hardware.

The startup did not disclose which chips were used to train the V3.1 model or which local processors the new FP8 format will support. Still, its comments hint at closer integration with China’s growing AI chip ecosystem, a move seen as critical for the country’s ambitions of technological self-sufficiency.