Anthropic Rejects Pentagon AI Deal

The Pentagon's decision to designate Anthropic a "supply chain risk" is an unprecedented use of a authority previously reserved for foreign adversaries like Huawei and Kaspersky. Grounded in 10 U.S.C. § 3252, the designation legally permits the DoD to exclude a company's products from defense contracts, but Secretary of Defense Pete Hegseth's claim that it bars all commercial activity between military contractors and Anthropic is being challenged as a statutory overreach. This high-stakes contracting environment is fueling a talent migration in Silicon Valley, with engineers and executives leaving consumer tech for defense-focused startups. Venture capital has poured over $28 billion into defense tech in 2025 alone, drawn by a nearly $1 trillion defense budget and a strategic shift to counter near-peer adversaries. This creates a competitive and complex talent retention market for established Bay Area tech firms. The underlying hardware for these large-scale AI contracts relies on a geopolitically sensitive semiconductor supply chain, with over 90% of the most advanced AI chips manufactured by TSMC in Taiwan. In response, the DoD is actively investing in domestic capabilities, including a $49 million initiative called RESHAPE to revitalize secure, advanced semiconductor packaging in the U.S. For military applications, AI hardware acceleration is focused on both massive cloud infrastructure and ruggedized edge computing. While the big models run in the cloud, there is a growing demand for on-device AI for drones, vehicles, and battlefield sensors that can operate without a constant cloud connection, driving development in low-power, high-performance custom silicon. San Francisco-based startup TurbineOne recently secured a $98.9 million contract for this type of edge AI. Google, also a recipient of a $200 million Pentagon contract alongside OpenAI and xAI, will provide its Cloud Tensor Processing Units for training AI models, representing a key hardware competitor in the defense space. The use of specialized hardware like Google's TPUs and rugged versions of NVIDIA's GPUs highlights a critical trend toward tailored hardware solutions for defense, moving beyond generic compute. This aggressive push into AI is part of a foundational shift within the DoD, building on programs like Project Maven, which uses computer vision to analyze drone and satellite imagery. That project, now integrated by companies like Palantir, transitioned from a controversial experiment that saw Google withdraw after employee protests to a core component of U.S. intelligence and targeting workflows.