Consent Didn't Disappear on Military Land. It Moved to the Courtroom.

The appeal of the military base was never really about the land. It was about jurisdiction. A data center on a base sits on federal property, where the zoning boards that have been blocking projects across the country have no authority ^[1]. Rep. Mark McCormick said the quiet part out loud: bases were a way to mitigate community opposition and land-use conflicts ^[2]. The bypass was sold as the place where the consent wall simply does not exist. The physical route is now real. Trump designated data centers critical national security infrastructure, reframing a land-use fight as a defense imperative ^[3]. The Army conditionally selected Carlyle for Fort Bliss and CyrusOne for Dugway Proving Ground under the Enhanced Use Lease program, under which the military receives computing power in exchange for the land ^[4]. The Pentagon is soliciting developers for more sites, treating the centers as classified military infrastructure under Pete Hegseth's "AI-first warfighting" memo ^[5]. On paper, the consent problem is solved: no zoning board, no public hearing, no moratorium. But the base is not just a landlord. It is also the customer, and the mission-setter. And the Pentagon's AI strategy has a second control point: what the AI is used for. That control point surfaced in a $200 million contract fight with Anthropic ^[6]. Claude's safety guardrails prohibited use for autonomous weapons targeting and mass domestic surveillance. Hegseth made the Pentagon's position explicit.

won’t allow you to fight wars — Pete Hegseth

Anthropic's Dario Amodei refused to remove them.

in all ways except those which would make us more like our autocratic adversaries. — Dario Amodei

The response came through the same federal toolkit that had cleared the land. Trump blacklisted Anthropic as a national security supply-chain risk, ordered federal agencies to stop using Claude, and Hegseth barred military contractors from doing business with the firm ^[7]. The Pentagon's classified AI work shifted to OpenAI and xAI, which complied with the demand for unrestricted access ^[7]. The instruments used to route around local consent — executive orders, national-security designations, blacklists — were now aimed at a company that had said no to a use case. The aftershock spread beyond Anthropic. Figma, Tenable, and Freightos warned investors that losing access to Claude could materially harm their businesses ^[8]. A single federal designation could cut off any company that depends on a model the military wants to use without limits. The consent question had become systemic. Then a federal judge stepped in. A court issued a preliminary injunction blocking the Pentagon's supply-chain risk designation of Anthropic, halting enforcement while the litigation continues ^[9]. The military could not simply compel access. The second consent wall — the one about what the AI is used for — turned out to be contestable, but only in a courtroom. The industry is now caught between two consent walls: the town that won't let it build, and the military that won't let it set limits on use. Most firms chose the wall they could live with. OpenAI and xAI took the military's terms ^[7]. Anthropic chose the one it couldn't, and only a judge could sort it out.