DeepSeek withholds new AI model from Nvidia
US officials have alleged that DeepSeek trained V4 using Nvidia’s advanced Blackwell chips at a facility in Inner Mongolia.
Salar News
San francisco/Singapore, 26 feb
Chinese artificial intelligence firm DeepSeek has withheld
pre-release access to its forthcoming V4 model from leading US chipmakers. This
fuelled speculation that Beijing is seeking to sideline American hardware
providers.
According to Reuters, DeepSeek did not share early versions of the model with Nvidia or Advanced Micro Devices (AMD), despite previously
collaborating with them to optimise performance on widely used processors.
Instead, early access was granted to domestic suppliers, including Huawei Technologies,
giving Chinese chipmakers a head start of several weeks.
AI developers typically provide hardware partners with
advance access to major updates to ensure software runs efficiently on their
systems. US officials have alleged that DeepSeek trained V4 using Nvidia’s
advanced Blackwell chips at a facility in Inner Mongolia, and may attempt to
obscure evidence of American chip usage while publicly claiming the model runs
on Huawei processors.
The model is expected to be released within weeks.
Analysts view the move as part of a broader Chinese strategy to disadvantage US hardware and AI models domestically while strengthening
national champions.
Since emerging in January 2025, DeepSeek’s models have been
downloaded more than 75 million times, with Chinese open-source offerings
gaining ground globally. DeepSeek’s debut triggered a historic market rout,
wiping between $465 billion and $600 billion from Nvidia’s market value in a
single day after claims it built world-class models for under $6 million.
Amid debate over export controls, US President Donald Trump recently said Nvidia could sell chips to approved Chinese buyers, subject to a 25 per cent US government fee.
Leave a Reply
Your email address will not be published. Required fields are marked *




