# CoreWeave Pushes Its AI Cloud Into Classified Data Centers > Source: > Published: 2026-07-30 12:23:01+00:00 ### [ Partnerships ](https://www.unite.ai/series/partnerships/) # CoreWeave Pushes Its AI Cloud Into Classified Data Centers [Add Unite.AI to your preferred sources on Google](https://www.google.com/preferences/source?q=unite.ai) CoreWeave ([CRWV](https://www.securities.io/nasdaq/CRWV/) ) and Leidos said on July 30, 2026 that they will [team up to deliver secure AI cloud services](https://www.coreweave.com/news/leidos-and-coreweave-collaborate-to-accelerate-delivery-of-ai-capabilities-for-defense-national-security-and-intelligence-missions) for U.S. intelligence and defense agencies, with CoreWeave’s platform running inside data centers accredited to handle classified material and Leidos owning the integration work around it. The division of labor is specific. CoreWeave plans to supply purpose-built infrastructure, high-speed networking, AI-optimized storage and cloud-native orchestration for both training and inference workloads, placed within Sensitive Compartmented Information Facilities, the accredited spaces where classified intelligence is stored and worked on. [Leidos](https://www.leidos.com/capabilities/ai) will lead mission integration, secure architecture accreditation support, cyber operations, data engineering and program delivery for the intelligence community and the Department of War. Both companies describe the product as sovereign AI capacity: compute dedicated to U.S. government workloads, built from the same stack CoreWeave rents to commercial AI labs, then adapted to classification and operational rules those labs never encounter. “This is the next evolution of mission technology—sovereign AI compute at scale, secure by design, mission integrated, operationally resilient, and ready for the realities of classified national security work,” said Jason O’Connor, president of Leidos Intelligence. ## What running GPUs in accredited space requires Accreditation, not silicon, is the constraint. Classified AI capacity has to sit in a facility cleared for the data, staffed by cleared people, isolated from the public internet, with auditability and model governance built in rather than bolted on. A commercial AI cloud is engineered for the opposite conditions: large multi-tenant regions, elastic allocation, and customers who can reach their clusters over the open network. That gap is what an integrator is being brought in to close, and it is why capacity of this kind cannot simply be carved out of an existing region. The two companies named five capability areas they intend to offer: **Classified AI cloud services** for model training, fine-tuning, evaluation, deployment and continuous monitoring**Analyst augmentation**, including multi-source intelligence fusion, imagery exploitation and decision support** Cyber AI ranges**for simulation, autonomous defense testing, threat modeling and adversarial AI evaluation** Synthetic data and simulation**, covering mission-relevant data generation and digital twins** Edge-to-cloud orchestration**, linking a central AI platform to forward-deployed and disconnected environments The last one is the hardest engineering problem in the set. Tactical environments have intermittent bandwidth and no tolerance for a round trip to a distant data center, which pushes inference onto local hardware and turns the central cloud into a training and update pipeline rather than a serving endpoint. Leidos brings the accreditation and program muscle for that work. It reported roughly $17.2 billion in revenue for the fiscal year ended January 2, 2026, employs more than 50,000 people, and already markets an on-premises AI platform aimed at federal agencies whose security requirements stall cloud deployments. Its own materials list sovereign compute as a focus area, alongside domestic talent and industrial capacity. ## Why federal demand matters to CoreWeave’s book CoreWeave’s commercial business is enormous and highly concentrated. In its [first-quarter results](https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-First-Quarter-2026-Results/) for the period ended March 31, 2026, the company reported $2.08 billion in revenue, up 112% year over year, and revenue backlog of $99.4 billion. Much of that sits with a handful of counterparties, including a $21 billion Meta commitment signed in March 2026 and a multi-year agreement with Anthropic. Federal work diversifies that mix, and it arrives on an entirely different cadence. Commercial capacity is bought with signed multi-year contracts that fund the buildout in advance; classified capacity moves at the speed of accreditation and appropriation. The companies said compute deployment and architecture will be determined by mission requirements and federal appropriation priorities, which is the honest description of how this scales. The capacity is there to draw on. CoreWeave passed 1 GW of active power in the first quarter and expanded contracted power past 3.5 GW, with chief executive Michael Intrator telling investors the company is on its way to more than 8 GW by 2030. Classified capacity will be a thin slice of that, constrained less by megawatts than by where accredited racks can legally live. The federal push has been building for some time. CoreWeave [announced its federal business](https://www.coreweave.com/news/coreweave-to-enter-the-u-s-federal-market) on October 28, 2025, aligning its platform with FedRAMP and other authorizations and staffing up in Washington. Agencies, meanwhile, have been buying commercial AI on aggressive terms, as the [$1 ChatGPT deal](https://www.unite.ai/u-s-federal-government-strikes-1-ai-deal-with-openai-to-expand-chatgpt-access/) showed, and hardware vendors have moved directly into defense programs, including [Nvidia’s Blackwell Ultra supercomputer for the Navy](https://www.unite.ai/nvidia-gives-the-navy-a-blackwell-ultra-supercomputer/). Sovereign capacity has become a stated procurement goal well beyond the U.S., with [Canada’s national AI strategy](https://www.unite.ai/canada-launches-national-ai-strategy-to-build-sovereign-infrastructure-scale-startups-and-drive-adoption/) among the clearest examples. ## What comes next Definitive agreements are the next step, which CoreWeave told investors the arrangement depends on. The commercial logic for both sides is already visible: Leidos gets a frontier-grade compute platform it does not have to build, and CoreWeave gets a demand channel that is not another AI lab, a move up the stack that echoes rival neoclouds like [Nscale’s purchase of Anyscale](https://www.unite.ai/nscale-buys-anyscale-to-move-up-the-ai-compute-stack/). For agencies, the offer is a commercial AI stack inside accredited space instead of a bespoke government build, and the two companies have named the capability areas they intend to sell against first.