The Pentagon's decision to bar Anthropic from its networks — and then quietly reach agreements with seven other AI companies to deploy on classified systems — looked like a startup moment. Generals calling small AI firms. Investors circling. The old guard scrambling.
But watch where the money actually flows, and a different story emerges. Legacy contractors aren't being displaced by this shift. They're positioning themselves as the indispensable layer between the Pentagon's AI ambitions and the startups trying to fulfill them.
The Infrastructure Play Nobody Talks About
The Anthropic saga revealed something important about how AI actually gets deployed at the DoD. When Claude was running on classified networks, it wasn't because Anthropic had a direct pipeline to the Pentagon. It worked because Palantir and Amazon Web Services built the secure infrastructure that made it possible — the platforms and cloud services that host AI models where they can touch classified data without leaking it.
That's the structural reality of defense AI: the model is the warplane, but the infrastructure is the runway. And right now, the runways are controlled by a small number of established players with the security clearances, compliance frameworks, and existing government relationships to operate them.
Nicolas Chaillan, founder of AI platform Ask Sage — used by thousands of DoD teams — estimates this secure AI infrastructure market at roughly $2 billion. That's not the AI model market. That's just the pipes. And the pipes are where legacy contractors and cloud hyperscalers are quietly consolidating their position.
The Partnership Pattern Is a Moat Strategy
When a defense prime announces an AI partnership with a venture-backed startup, the framing is usually about innovation — the old guard embracing new thinking. I'd argue the more accurate framing is defensive positioning. By partnering early, primes get to shape how a startup's technology integrates with existing systems, which contracts it pursues, and which security frameworks it operates within.
The Anthropic ouster opened real doors for smaller AI rivals — generals and combatant commanders suddenly fielding calls from companies they'd never heard of. But fielding calls is not the same as winning contracts. The startups that convert that attention into revenue are the ones that can clear the security and integration hurdles fast enough to matter. Those hurdles favor companies that already have established infrastructure partners — which means, in practice, companies that have already made peace with the primes.
This isn't cynicism about the startups. It's a structural observation: the DoD's classified network requirements create a natural chokepoint, and the companies controlling that chokepoint have enormous leverage over which AI capabilities actually reach operational use.
What the True Anomaly Round Actually Signals
Space security startup True Anomaly raised $600 million this week, leading what Crunchbase identified as the week's largest funding rounds — with defense tech accounting for multiple large deals. That's a significant capital signal, but the more interesting question is what True Anomaly does with it.
Space security is a domain where the infrastructure dependency problem is acute. Satellites, ground stations, and command networks are deeply integrated with legacy systems. A startup with $600 million and genuinely differentiated technology still has to find a path through the same classified network requirements, the same integration timelines, the same procurement frameworks that every other defense tech company navigates.
The ones that figure this out — that find the right infrastructure partner, clear the right certifications, and get embedded in the right program offices — become the next generation of primes. The ones that don't become cautionary tales about how hard it is to sell to the Pentagon even when the Pentagon wants what you're building.
The Dcode Model Points at the Real Problem
Meagan Metzger founded Dcode in 2015 after watching the same pattern repeat: government building software it could buy, commercial tech firms refusing to touch the federal market, and the gap between them widening. Her insight — that the technology was never the hard part, that "it's all the unsexy stuff that gets in the way" — is still the most accurate description of why defense tech transformation moves slowly.
The AI partnership wave between legacy contractors and startups is, in part, an attempt to solve that unsexy stuff at scale. Whether it actually accelerates capability delivery or just creates a new layer of institutional friction is the question worth watching over the next 18 months — particularly as the Pentagon's seven new AI agreements move from announcement to actual deployment on classified networks.
Watch for which of those seven companies announces infrastructure partnerships with established primes. That's the tell.