What the Defense Production Act Can and Can’t Do to Anthropic

On Feb. 24 Defense Secretary Pete Hegseth warned Anthropic CEO Dario Amodei he would invoke the Defense Production Act (likely Title I) unless Anthropic accepts Pentagon terms, pressing for Claude to be supplied without contractual guardrails or even retrained to remove safety limits. Legal analysts say the DPA's priority-access versus compelled-contracting/allocation powers are ambiguous—demands to change access terms may be defensible while forced retraining looks more like requiring a new product and raises major-questions and possible First Amendment issues—making litigation likely and underscoring the lack of Congressional rules for military AI.

Analysis

Market structure: A credible DPA threat is a net positive for large, incumbent defense primes (RTX, LMT, NOC) and hyperscaler clouds (MSFT, AMZN, GOOGL) that can absorb classified/regulated workloads and provide secure hosting. Smaller pure-play model vendors and high-multiple AI SaaS names lose pricing power and face forced repricing as government demand (potentially 1–3% of commercial model capacity) re-routes to defense channels over 12–18 months. Expect contract terms to tilt toward fixed-price, compliance-heavy deals, compressing gross margins for vendors without scale or Fed contracting experience.

Risk assessment: Tail risks include a forced-retraining DPA order (20–35% probability in next 3 months) that would trigger litigation, brand damage, and possible injunctions; courts may block extreme compulsion (10–20% chance of successful injunction within 30–90 days). Immediate (days) volatility will be headline-driven; short term (weeks–months) sees reallocation into defense/cloud; long term (6–18 months) depends on Congress — a clarifying law either codifying or narrowing DPA use is the single largest systemic catalyst. Hidden dependencies: classified accreditations, cloud region capacity, and talent retention (engineers leaving companies that cooperate with military) can shift outcomes materially.

Trade implications: Prefer quality incumbents: allocate to defense primes and hyperscalers that already have Fed footholds; size initial entries modestly (1.5–3% each) and scale if DPA is invoked or classified RFPs increase by >20% YoY. Use options to express view: buy 3–6 month call spreads on RTX/LMT (5–10% OTM) to limit capital at risk, and buy short-dated puts on high-multiple AI pure-plays to hedge headline shocks. Rotate out of small/mid-cap pure AI SaaS into cybersecurity (PANW, FTNT) and cloud infra over 4–12 weeks as bids for defense work firm up.

Contrarian angles: Consensus treats regulatory compulsion as purely negative for all AI players; investors miss that government backing (via DPA or contracts) creates durable revenue streams and higher barriers to entry benefiting incumbents — think 10–20% revenue stickiness over 2 years. Historical parallels: post-9/11 security spend and Cold War-era industrial consolidation show that conditional government demand can re-rate defense/cloud incumbents while culling startups. Unintended consequences include consolidation opportunities (M&A) among distressed model vendors and a legislative push within 6–12 months that could either entrench or sharply limit DPA reach, which would pivot winners/losers quickly.

Key Decisions for Investors

• Establish a 2–3% long position in RTX and a 2% long in LMT within the next 10 trading days; target 12–18% upside over 12 months, set stop-loss at 8% below entry to limit drawdown if litigation/injunctions reduce contract flow.
• Add 1.5–2% long positions in MSFT and AMZN focused on Azure/AWS government cloud exposure; express leverage via 3–6 month call spreads (buy ATM, sell 7–10% OTM) to cap cost while targeting 10–20% move if classified/regulated workloads accelerate.
• Reduce aggregate exposure to small/mid-cap pure-play AI SaaS names by 40–60% within 30 days (trim positions >5% of portfolio to 2–3%) and redeploy into defense/cloud/cybersecurity, as margin compression and legal risk reprice multiples.
• Purchase 0.5–1% of portfolio in 3–6 month out-of-the-money puts on a representative high-multiple AI large-cap (e.g., NVDA) as tail protection; initiate if NVDA gaps down >8% on DPA headline within 5 trading days or if courts permit forced retraining.
• If Congress files clarifying legislation within 90 days that limits DPA compulsion for AI (watch bill text and committee calendar), shift 1–2% from defense incumbents back into AI growth leaders (NVDA, GOOG) within 30 days of bill passage/score to capture rerating.