NVIDIA, Microsoft and AWS win Pentagon approval to run advanced AI on classified networks

The Pentagon just moved a big chunk of the AI race behind the classified wall. It signed new agreements that let seven companies — NVIDIA, Microsoft, AWS, Google, OpenAI, SpaceX, and Reflection — put advanced AI tools onto Defense Department classified networks. That sounds bureaucratic, but it is the part that actually matters. Fancy models are one thing. Getting them approved to touch Secret and Top Secret workflows is the real gate. (nextgov.com) ### What changed? The new step is not “the Pentagon likes AI.” That was already obvious. The change is that these vendors can now support work inside Impact Level 6 and Impact Level 7 environments — the classified network tiers the Defense Department uses for Secret and higher-end national security workloads. The P(nextgov.com)ional awareness. (nextgov.com) ### Why are IL6 and IL7 the real story? Because classified access is the bottleneck. Plenty of companies can demo an AI model on an unclassified laptop. Far fewer can operate inside the security, isolation, compliance, and cleared-support requirements that come with military classified systems. AWS’s Secret cloud i(nextgov.com)is the plumbing these approvals sit on top of. (aws.amazon.com) ### Where do Microsoft and NVIDIA fit? Microsoft already had a key piece in place. Azure OpenAI Service was authorized in 2025 for all U.S. government data classification levels, including Impact Level 6 and Top Secret environments. So this Pentagon move is less about proving Microsoft can handle classified AI and more about routing those capabilities into a broader Defense workflow. NVIDIA is (aws.amazon.com)s, inference stack, and model-serving ecosystem that make a lot of frontier AI usable at scale. (devblogs.microsoft.com) ### Why include so many vendors at once? Basically, the Pentagon is trying to avoid getting trapped by one supplier. The department said outright that the goal is to build an architecture that prevents vendor lock and preserves long-term flexibility. That is notable because defense procurement usually drifts toward entrenched incumbents. Here, the Pentagon is sign(devblogs.microsoft.com)the technology changes. (nextgov.com) ### So is this really about cloud contracts? A lot of it is. Once a model is approved for classified use, the spending does not stop at the model API. It pulls in GPUs, secure cloud capacity, storage, networking, identity controls, monitoring, and cleared support staff. In other words, “AI approval” often turns into a much larger infrastructure relationship. That is why AWS, Microsoft, and NVIDIA all matter here even though they play different roles. (aws.amazon.com) ### Why is Anthropic missing? That absence is part of the story. The Pentagon’s latest classified AI push comes after a public rupture with Anthropic over military use restrictions. The company is not in this new group, while rivals are. So the Defense Department is not just expanding AI access — it is also reshaping which vendors get to become embedded in national security workflows. (nextgov([aws.amazon.com)ntagon-makes-agreements-7-companies-add-ai-classified-networks/413264/)) ### What does this unlock next? It means defense users can start building habits around these tools inside the environments that matter most. Once analysts, operators, and program offices rely on a model inside classified systems, that vendor gains a much stronger position for follow-on work. The catch is that cybe(nextgov.com)not waved through because the model is impressive. (dodcio.defense.gov) ### Bottom line? This is not just a Pentagon AI experiment. It is a permissions change at the classified edge of government computing. And in defense tech, permissions are power. (nextgov.com)