The Department of Homeland Security posted a notice instructing civilian federal agencies to take DeepSeek products off their networks. A companion directive dated July 24, signed by Markwayne Mullin, also shuts the company out of federal procurement.
The directive says DeepSeek gathers large volumes of user data that could be shared with Beijing under Chinese information security rules. It references a 2018 statute designed to tighten security in government purchasing. Embassy representatives for China in Washington did not immediately comment.
DeepSeek vaulted into view when its R1 model jolted global markets early last year, drawing attention from U.S. policymakers who see national security risks. In April 2025, the House Select Committee on China said the firm's connections to Beijing are "significant," and lawmakers proposed legislation to mandate removal of its software from federal systems.
Some agencies were already acting. In January 2025, the Pentagon's Defense Information Systems Agency took steps to cut off access to the company's chatbot. The Navy already advises against using open source AI for official tasks.
Several U.S. states have also put limits on its use. U.S. national security officials have accused DeepSeek and other leading Chinese AI developers of systematically pulling proprietary know how from American companies. In February, OpenAI told U.S. lawmakers that DeepSeek used a technique called distillation to tap leading U.S. AI models while creating its R1 chatbot.
When policy and tech developments affect markets, stay grounded with free Market Briefs that explain it in five minutes or less
U.S. officials worry about dependence on Chinese AI models like DeepSeek's, yet say policy choices have been constrained by a shortage of domestic substitutes for lower cost, customizable open weight models that let users download the core tech. That is starting to change. Earlier this month, Reflection AI announced Beam, its first open weight model, claiming it can match Chinese options.
Nvidia has been expanding its Nemotron line of open models. These U.S. offerings are also vital for edge computing - processing information close to where it is generated - a capability that underpins certain government and military missions.
Meanwhile, American firms such as OpenAI face stronger competition from Chinese players - among them DeepSeek and Moonshot AI - that sell lower-cost open-weight models.
Beijing recently put forward a plan to add advanced AI technologies to its export control catalog, which could restrict sending powerful models abroad. The draft lists 23 AI categories, from large language models and agents to chip architecture and compute clusters. Items that cross certain performance bars would be limited, including the model weights and source code.
The bottom line: the U.S. is tightening the gate on Chinese AI inside federal systems while trying to seed homegrown open models. That mix could steer demand toward American open weight offerings, lift spend on edge computing gear, and add uncertainty for vendors exposed to Chinese AI supply chains. Watch how contracting shifts and export rules evolve - they set the lanes where capital can flow.
Big policy shifts can reshape investor attention, so sign up for our free Market Briefs newsletter for concise daily context
