Chinese DeepSeek withholds latest AI model from US chipmakers

(Patria) - DeepSeek, a Chinese artificial intelligence (AI) lab whose inexpensive model shook global markets last year, has not shown its upcoming flagship performance optimization model to U.S. chipmakers, two people familiar with the matter said, breaking standard industry practice ahead of a major model update, Reuters reported.

Instead, the lab, which is expected to launch its next major update, V4, has granted early access to domestic suppliers, including Huawei Technologies, the sources said.

AI developers typically share pre-release versions of major models with leading chipmakers such as Nvidia and Advanced Micro Devices (AMD) to ensure their software runs efficiently on widely used hardware. DeepSeek has previously worked closely with Nvidia's technical staff.

Nvidia and AMD left out

For its upcoming model, which was expected around the Lunar New Year holiday, DeepSeek did not grant access to Nvidia and AMD and gave Chinese chipmakers, including Huawei, a head start of several weeks to optimize software for their processors, the sources said.

Nvidia and AMD declined to comment. DeepSeek and Huawei did not respond to Reuters requests for comment. Reuters could not immediately determine the reason for the decision.

"The impact on Nvidia and AMD in terms of general data accelerators is minimal. Most enterprises do not use DeepSeek which serves as a benchmark model more than anything else," said Ben Bajarin, chief executive of research firm Creative Strategies, adding that new AI coding tools reduce the time it takes for software to run well on hardware, "from months to weeks."

The move is likely part of a broader Chinese government strategy "to try to disadvantage U.S. hardware and models" in China, Bajarin said.

This comes after a senior official in the Donald Trump administration told Reuters that DeepSeek's latest AI model was trained on Nvidia's most advanced chip, Blackwell, using a cluster in mainland China, in a move that appears to violate U.S. export controls.

Was the law broken?

DeepSeek may try to remove technical indicators that reveal its use of U.S. AI chips and plans to publicly claim it used Huawei's chips to train its model, according to a U.S. official.

DeepSeek's models have been downloaded more than 75 million times on the open-source platform Hugging Face since the company emerged on the scene in January 2025, contributing to the rise of Chinese open-source models competing with U.S. AI labs. Among models released in the past year, downloads of Chinese models have surpassed those from any other country on the platform.

The rapid rise of Chinese open-source models has intensified debate in Washington about exporting advanced U.S. AI chips to China. U.S. authorities last year 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 remain restricted. It was unclear whether DeepSeek had secured approval to purchase those U.S. chips.

The H20 and MI308 chips are intended for inference, the process of running trained AI models. Demand for the MI308 has been significant, with AMD reporting $390 million in chip sales in the last quarter.

DeepSeek is among several Chinese AI firms expected to introduce new models this month.