DeepSeek hides upcoming flagship AI model from US chipmakers like Nvidia, shows preview to Chinese Huawei: Report

DeepSeek, the Chinese artificial intelligence laboratory whose low-cost model shook global markets last year, did not show US chipmakers its upcoming flagship model for performance optimization, two sources said, breaking with standard industry practices before a major model update.

The lab, which is expected to release its next major update, V4, instead gave early access to domestic suppliers, including Huawei Technologies, sources said.

AI developers often share pre-release versions of major models with leading chipmakers such as Nvidia and Advanced Micro Devices to ensure their software performs efficiently on commonly used hardware. DeepSeek has previously worked closely with Nvidia’s technical team.

Nvidia and AMD were left out

Sources said DeepSeek did not provide access to Nvidia and AMD for its new model, which is expected to be released around the Lunar New Year holiday, giving Chinese chipmakers, including Huawei, a few weeks’ head start to optimize the software for their processors.

Nvidia and AMD declined to comment. DeepSeek and Huawei did not respond to requests for comment.

Reuters could not immediately determine the reason for the decision.

“Nvidia and AMD’s impact on mainstream data accelerators is minimal; most organizations do not run DeepSeek, which serves as a benchmark model more than anything else,” said Ben Bajarin, CEO of research firm Creative Strategies. He added that new AI coding tools have reduced the time needed to ensure software runs well on hardware from “months to weeks.”

The move is likely part of a broader strategy by the Chinese government “to keep U.S. hardware and models at a disadvantage in China,” Bajarin said.

The development comes after a senior Trump administration official told Reuters DeepSeek’s latest AI model was trained on Nvidia’s most advanced chip, Blackwell, using a cluster in mainland China, in a move that appeared to violate US export controls.

According to the US official, DeepSeek may seek to remove technical indicators that reveal it is using American AI chips and may plan to publicly claim that it used Huawei’s chips to train its model.

DeepSeek’s models have been downloaded more than 75 million times on the Hugging Face open-source platform since the company burst onto the scene in January 2025, helping fuel a wave of Chinese open-source models competing with U.S. AI labs. Among the models released last year, the number of downloads for Chinese models surpassed downloads in other countries on the platform.

The rapid rise of Chinese open source models has intensified debate in Washington over exporting advanced US AI chips to China. Last year, US officials allowed Nvidia’s H20 and AMD’s MI308 chips designed for AI inference to continue shipping to China, even as licenses for more advanced processors remained restricted. It was unclear whether DeepSeek had received approval to purchase these US chips.

The H20 and MI308 chips target inference, the process of running trained AI models. Demand for MI308 was significant; AMD announced that it earned $390 million from chip sales in the last quarter.

DeepSeek is among several Chinese artificial intelligence companies expected to announce new models this month.