The U.S. Department of Defense has disclosed the seven companies awarded AI procurement contracts. The list includes Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, SpaceX, and the relatively little-known startup Reflection AI; Anthropic was left off the list.
The awards expand services available through the department’s internal AI portal, GenAI.mil. According to reports, more than 1.3 million Defense Department personnel have used the platform since its launch last year, and it now hosts hundreds of thousands of AI agents. The department says the portal is intended to reduce data-aggregation tasks and accelerate decision-making.
Deployed in highest-grade environments that handle classified information
The AI products covered by the new contracts are slated to run in Impact Level 6 and Impact Level 7 environments. Those designations apply to Defense systems that can store classified information and reflect the most stringent security requirements. The department has not disclosed which specific products were selected, but the procurement marks a clear move from experimental testing to classified-level AI operations.
Analysts are also watching what each company can contribute. NVIDIA, best known for graphics processing units (GPUs), also offers neural-network development tools and open-source large language models. Models built on the Mamba-Transformer architecture, which reduce memory use, could be attractive for defense use cases that operate in constrained compute environments.
SpaceX and Reflection AI draw attention; Anthropic excluded
SpaceX’s inclusion is notable. After combining forces with xAI Holdings this year, SpaceX gained access to the Grok family of language models and has been raising its profile as an AI supplier. Observers say an expanded portfolio of coding-focused models could accelerate the company’s transition from aerospace builder to defense AI vendor.
Reflection AI is the most unfamiliar name on the roster. Founded in 2024 by researchers from Google DeepMind, the startup has not released a commercial product but reportedly raised $2 billion last year (approximately 2.67 trillion KRW). U.S. media say the company is preparing a language model trained on tens of trillions of tokens.
By contrast, Anthropic’s exclusion was largely anticipated. Defense Secretary Pete Hegseth designated Anthropic a supply-chain risk in February, and the department formalized that designation in March. The decision led to a ban on using Anthropic’s Claude models within the department and restricted access for defense contractors.
The Claude ban stemmed from a contractual dispute over “all lawful purposes” language
The dispute centered on contract language. Secretary Hegseth said he imposed the restriction after Anthropic declined the department’s requirement that its models be usable for “all lawful purposes.” Anthropic argued that wording could open the door to uses such as large-scale domestic surveillance or autonomous weapons development. In March, the company filed suit to contest its supply-chain risk designation.
The reality is more complicated. Reports indicate the department has continued to use Anthropic’s Claude Mitos Preview despite the ban. The model has seen limited adoption at other federal agencies but is not publicly available. Sources say it is highly effective at uncovering previously unknown zero-day vulnerabilities, which makes it valuable for certain security missions.
These procurement moves signal more than technology adoption: they reveal which companies’ AI systems the national security establishment is willing to trust inside protected environments. The Anthropic episode also underscores that, in the defense market, performance alone is not enough—contract language and policy compliance are now central to who gains access to the AI supply chain.
TP AI Notice This article was summarized using a TokenPost.ai-based language model. Key details from the original may have been omitted or differ from the facts.
