A Chinese artificial intelligence company that shocked global markets last year with its cost-effective technology has now shut out major American chip manufacturers from accessing its newest AI model before its public release, according to two industry sources.
DeepSeek, the AI laboratory behind the disruptive low-cost model, has denied Nvidia and Advanced Micro Devices early preview access to its anticipated V4 update, departing from typical industry protocols, sources familiar with the situation revealed.
The Chinese firm has instead granted domestic technology companies, including Huawei Technologies, exclusive early access to the forthcoming model, giving them a multi-week advantage to fine-tune the software for their processing systems.
This approach contradicts established industry standards where AI companies routinely provide pre-launch versions of significant models to major chip manufacturers like Nvidia and AMD. This collaboration ensures optimal software performance on commonly used hardware platforms. DeepSeek had maintained close working relationships with Nvidia’s technical teams in previous projects.
The anticipated V4 model was originally scheduled for release during the Lunar New Year celebrations. While Chinese chip companies received weeks of advance optimization time, American manufacturers were completely excluded from the process, sources indicated.
Both Nvidia and AMD representatives declined to provide statements regarding the situation. DeepSeek and Huawei did not respond to inquiries seeking comment.
The specific reasoning behind DeepSeek’s decision remains unclear, according to available information.
Ben Bajarin, CEO of Creative Strategies research firm, assessed the situation’s impact, stating: “The impact to Nvidia and AMD for general data accelerators is minimal – most enterprises are not running DeepSeek, which serves as a benchmarking model more than anything else.” He noted that emerging AI programming tools are accelerating software-hardware optimization timelines “from months to weeks.”
Bajarin suggested this decision likely reflects a wider Chinese government initiative “to try to keep U.S. hardware and models disadvantaged” within China’s market.
These developments coincide with revelations from a senior Trump administration official who told reporters that DeepSeek’s newest AI system was developed using Nvidia’s cutting-edge Blackwell processor through a mainland China-based cluster, potentially violating American export restrictions.
According to the US official, DeepSeek may attempt to conceal technical evidence of American AI chip usage and publicly assert that Huawei processors were used for model training.
Since DeepSeek emerged prominently in January 2025, its models have been downloaded over 75 million times through the Hugging Face open-source platform, contributing to a surge of Chinese open-source models challenging American AI laboratories. Chinese model downloads have exceeded those from all other nations on the platform among models released in the past year.
The growing prominence of Chinese open-source AI models has heightened Washington discussions regarding advanced US AI chip exports to China. Last year, US officials permitted resumed shipments of Nvidia’s H20 and AMD’s MI308 processors – designed for AI inference operations – to China, while maintaining restrictions on more sophisticated processors. Whether DeepSeek has obtained authorization to purchase these American chips remains uncertain.
The H20 and MI308 processors focus on inference operations, which involve running completed AI models. The MI308 generated substantial demand, with AMD reporting $390 million in chip sales during its latest quarterly period.
DeepSeek joins several other Chinese AI companies planning to introduce new models this month.
