Pentagon tests AI rivals as Anthropic faces legal and chip-access pressure—who wins the next model race?

On May 19, the U.S. D.C. Court of Appeals heard oral arguments in a lawsuit filed by Anthropic against Secretary of Defense Pete Hegseth and the Department of Defense, according to Lawfare. In parallel, Bloomberg reports that the Pentagon is testing rival AI models to determine which are favored by 25 “power users” inside the department as the U.S. military seeks alternatives to Anthropic PBC’s Claude. Separately, The Information reports that Anthropic is in talks to use Microsoft’s AI chips, signaling a potential shift in compute sourcing and deployment strategy. Taken together, the cluster suggests Anthropic is navigating both legal constraints and supply-chain/compute negotiations while the Pentagon accelerates model diversification. Strategically, this is a competition over who controls the most trusted language-model layer inside U.S. defense workflows—an issue that blends procurement, operational security, and technological sovereignty. The Pentagon’s internal “power user” evaluation implies a move away from single-vendor dependence, which can reduce risk from licensing, performance variability, or policy/legal exposure. Anthropic’s reported talks with Microsoft point to a balancing act: maintaining access to leading-edge inference/training capacity while staying operationally usable for government customers. The likely beneficiaries are firms that can demonstrate defensible performance, compliance posture, and reliable access to advanced chips, while the main losers are any vendors that become bottlenecked by legal uncertainty or compute constraints. Market and economic implications center on AI infrastructure and defense-linked technology spending rather than traditional commodities. If the Pentagon expands beyond Claude, demand could shift toward alternative model providers and toward the specific chip ecosystems used for inference at scale, affecting sentiment around AI accelerators and cloud capacity. Microsoft’s AI chip supply discussions with Anthropic, if confirmed, could reinforce expectations for continued enterprise and government utilization of Microsoft’s hardware/software stack. In the near term, investors may watch for volatility in AI platform multiples tied to government contracts, as well as for second-order effects on cybersecurity and compliance tooling used to govern model behavior. While the articles do not quantify dollar amounts, the direction is clear: procurement diversification and compute sourcing negotiations are likely to intensify competition across AI model, cloud, and accelerator supply chains. What to watch next is the D.C. Circuit’s follow-on actions after the May 19 arguments, including any written opinions, injunction-related developments, or procedural rulings that could affect DoD’s ability to rely on specific products or contractual terms. On the Pentagon side, the key trigger is the outcome of the “25 power users” model testing—whether it yields a shortlist, a phased migration plan, or a formal procurement pathway. For Anthropic, the next milestone is whether the Microsoft chip talks translate into signed agreements, deployment timelines, or measurable performance improvements in government-relevant settings. Finally, broader signals—such as continued public controversy around AI adoption and the pace of AGI-focused research narratives—can influence policy scrutiny, which in turn can affect procurement timelines and compliance requirements across the defense AI stack.

Geopolitical Implications

• 01

  Defense AI procurement is becoming a strategic contest over model reliability, compliance, and compute access—effectively a form of technological sovereignty.

• 02

  Legal disputes can translate into operational risk management, pushing governments toward multi-vendor architectures and faster model substitution.

• 03

  Compute supply-chain negotiations may determine which AI providers can sustain government-grade performance and continuity.

Key Signals

• —D.C. Circuit written opinion or injunction-related developments after May 19.
• —Pentagon shortlist or phased migration plan based on the 25 “power users” tests.
• —Confirmation of Anthropic–Microsoft chip agreements and deployment timelines.
• —Any procurement documents referencing alternative models to Claude.