TL;DR — Key Takeaways
– Federal agencies must remove DeepSeek technology: The DHS order covers civilian agency systems, including third-party products built using DeepSeek.
– DeepSeek is opening its inference software: DeepInfer supports DeepSeek-V4.1-Flash on NVIDIA and Huawei hardware.
– Lower costs remain a deployment challenge: DeepInfer could improve inference efficiency, but enterprises need additional infrastructure to build a complete service.
The U.S. government and the artificial intelligence (AI) industry are facing a pivotal juncture as the federal government ramps up security restrictions on Chinese technology while open-source software continues to disrupt the global AI market.
The Trump administration formally barred federal civilian agencies from acquiring or utilizing technology developed by DeepSeek.
The directive, issued by the Department of Homeland Security (DHS) in late July and made public this week, requires federal civilian entities — including the Treasury, State, and Justice departments — to purge DeepSeek-related software from their information technology systems within 60 days.
The mandate applies to both direct software implementations and third-party commercial services built using DeepSeek’s models, though it excludes the Department of Defense.
In its rationale, the DHS cited significant national security risks, warning that deploying DeepSeek technology creates threat vectors through potential cybersecurity vulnerabilities, malware, and unauthorized access by Chinese authorities to sensitive American administrative data. The order permits federal executive agencies to submit written requests for compliance with delays or specific exceptions within a 60-day window.
U.S. restrictions have drawn pushback from Beijing. Chinese Foreign Ministry spokesperson Mao Ning said China consistently opposes the politicization of technology and trade issues under the guise of national security, emphasizing that Beijing will take necessary measures to protect the rights of its domestic firms.
“The DeepSeek decision reflects a growing reality facing federal agencies: AI adoption can’t outpace security and governance,” said Gary Barlet, Public Sector CTO at Illumio. “When agencies don’t have confidence in where their data is stored, who can access it, or how an AI model handles that information, restrictions like this become inevitable. The concern isn’t just one company or one model. It’s the broader challenge of ensuring sensitive government data doesn’t leave approved environments or create new pathways for adversaries to gain insight into government operations.”
The civilian ban follows earlier restrictions across the federal footprint, including congressional bans on intelligence systems and individual administrative blocks by the Commerce Department and the Defense Information Systems Agency.
Despite growing regulatory friction in Washington, D.C., DeepSeek is advancing its market strategy by releasing its internal production software into the open-source ecosystem. The startup has published DeepInfer, the underlying software framework used to execute its models, on GitHub under an open-source Apache 2.0 license.
DeepInfer currently supports DeepSeek-V4.1-Flash, a high-efficiency model designed to drastically lower operational costs. Built on top of popular open-source frameworks like vLLM, SGLang, and NVIDIA Corp.’s TensorRT-LLM, the software incorporates advanced optimization techniques such as speculative decoding and multi-chip workload distribution to accelerate output generation. DeepInfer is engineered to run on both NVIDIA SM100 data-center GPUs and Huawei Ascend 950 hardware.
The timing of the open-source release addresses a major financial bottleneck for AI deployment: inference costs. As enterprise workflows increasingly demand complex, multi-step agentic capabilities, running models in production has surpassed initial training as the dominant computing expense.
DeepSeek’s low token pricing — currently listed at fractions of a cent per million tokens — presents a stark contrast to higher-priced proprietary API offerings from American developers like OpenAI and Anthropic.
While the new federal ban isolates DeepSeek from U.S. government procurement, the open-source release of DeepInfer provides private commercial enterprises worldwide with direct access to lower compute costs, sharpening competition across the global technology sector.
