Chinese artificial intelligence startup DeepSeek has unveiled its updated large language model, V3.1, and hinted that China will soon see domestically produced “next generation” chips to power advanced AI systems.
In a comment on its official WeChat account, DeepSeek said the model’s “UE8M0 FP8” precision format was designed for compatibility with upcoming locally made chips. FP8, or 8-bit floating point, is a data format that increases efficiency in training and inference for large-scale deep learning.
Push for Domestic Alternatives
The statement comes as Beijing urges Chinese AI developers to reduce reliance on Nvidia and embrace domestic alternatives amid U.S. semiconductor export restrictions. Huawei and other local players have been building momentum, although analysts note they still lag Nvidia in both scale and capability.
DeepSeek did not specify which chips trained the new model or identify the local processors expected to work with the UE8M0 FP8 format.
Navigating U.S. Restrictions
Earlier this year, DeepSeek gained international attention when its R1 reasoning model matched the performance of Western systems, despite being trained without access to Nvidia’s most advanced chips due to U.S. export bans. Its earlier V3 release in December 2024 had been trained on roughly 2,000 less powerful Nvidia processors.
The U.S. tightened restrictions again in April, banning Nvidia’s H20 chips designed specifically for the Chinese market. Although the Trump administration later said it would allow some shipments to resume, Chinese regulators have reportedly instructed firms not to buy the chips pending a national security review.
DeepSeek described V3.1 as featuring faster response times and a hybrid reasoning architecture, capable of switching between reasoning and non-reasoning modes. Reasoning models enable step-by-step logical processes for more complex tasks.
The company also announced that starting September 6, pricing for its API access will be updated, allowing developers to integrate V3.1 into apps and platforms with revised cost structures.
What The Author Thinks
DeepSeek’s reference to “next-generation” local chips shows how critical semiconductor independence has become for China’s AI future. The country cannot afford to rely indefinitely on foreign suppliers, especially with U.S. export restrictions tightening every year. Even if Chinese chips fall short of Nvidia’s performance now, the sheer demand for AI tools in China ensures massive state and private investment. That pressure could push local firms to catch up faster than expected. DeepSeek aligning its models with domestic formats suggests it is betting on this inevitability—and if the bet pays off, it could reshape the global AI hardware race.
Featured image credit: Solen Feyissa via Unsplash
For more stories like it, click the +Follow button at the top of this page to follow us.