At a Glance
- OpenAI, Meta, Anthropic, and Google all dropped prohibitions on defense use of their AI models within a single year
- Defense contracts now span cloud, surveillance, and weapons development
- Why it matters: The fastest-growing tech sector is betting its future on Pentagon spending, not consumer markets
America’s largest AI labs spent 2024 dismantling every major safeguard against military use of artificial intelligence. The reversal was swift, coordinated, and driven by a shared need for the one customer with deep pockets and little patience: the U.S. Department of Defense.
The 180-Day Unwind
In January 2024, OpenAI quietly deleted the clause that barred “military and warfare” applications from its usage policies. By November, the firm confirmed “a number of projects” with the Pentagon. Meta followed within days, granting the U.S. and select allies access to its Llama models for defense purposes. Anthropic announced a partnership with Palantir the same week. Google capped the year by rewriting its AI principles to permit weapons work.
The dominoes fell in six months-a pace that startled even longtime employees.
| Company | Old Policy | New Policy | First Known Defense Deal |
|---|---|---|---|
| OpenAI | No military use | Allow defense projects | Pentagon “projects” (Jan 2024) |
| Meta | No weapons | Llama for U.S. + allies | November 2024 |
| Anthropic | No military use | Palantir partnership | November 2024 |
| No weapons | Principles rewritten | February 2025 |
Why They All Said Yes
The labs face the same economic squeeze. Training frontier models now costs $100 million-$1 billion per run, while commercial revenue lags. Defense contracts offer three things Silicon Valley craves:
- Multi-year, multi-billion-dollar deals with soft budget constraints
- Lenient success metrics compared with consumer products
- Early-adopter status that can lock out competitors for decades
Economist David J. Teece summarized the dynamic in 2018: “GPTs develop faster when there’s a large, demanding, and income-generating application sector-such as the U.S. Defense Department’s purchases of early transistors and microprocessors.”
From Consensus to Cage Match
The pivot ends the “Silicon Valley Consensus” that shaped U.S. tech policy from the 1990s through the mid-2010s. That era rested on three pillars:
- Globalized data flows would spread American influence
- Minimal regulation would maximize innovation
- Economic integration with China would liberalize Beijing
Washington and Big Tech moved in lockstep. Section 230 shielded platforms from liability; trade deals exported those protections; China got WTO entry in 2001. Even geopolitical crises-Taiwan missile tests, embassy bombings, spy-plane collisions-were smoothed over to keep supply chains intact.
The Fracture
The consensus cracked in three stages:
- 2016-2018: Trump tariffs signal the end of free-trade orthodoxy
- 2019-2021: Capitol Hill hearings, antitrust suits, and content-moderation battles turn tech CEOs into hostile witnesses
- 2022-2024: Ukraine war, China chip bans, and the AI arms race make national security the top brief
Two rival camps now fight for control of the U.S. tech-state alliance.
Big Tech Globalists
Amazon, Google, Microsoft still want frictionless data flows-they earn 60-70% of revenue overseas. But they now court spy agencies for both money and protection:
- Amazon signed CIA cloud deals in 2013, 2020, a $10 billion NSA contract in 2021, and an Army pact in 2024
- All cloud giants host classified regions for intel agencies
- OpenAI appointed former NSA director Paul Nakasone to its board in 2024
Tech-Nationalist Right
Startups such as Anduril, Palantir, and Shield AI market themselves as faster, cheaper heirs to Lockheed and Raytheon. They enjoy:
- Venture capital from Andreessen Horowitz and General Catalyst
- $14 trillion in Pentagon spending since 2001-one-third to just five legacy contractors they aim to unseat
- A White House that openly favors reshoring and arms races
Talking Like Hawks
Lab leaders now echo defense talking points. Sam Altman shifted from urging U.S.-China AI dialogue to writing in the Washington Post that the future of AI should sit with a “U.S.-led coalition of like-minded countries.” Anthropic CEO Dario Amodei warns of a “division between democratic countries and authoritarian ones” and insists “democracies must win.”
The rhetoric matches policy: export controls on advanced chips, investment screening for Chinese AI ventures, and tariffs on Chinese goods have all tightened since 2022.
What Disappeared
Existential-risk debates-once central to AI-safety discourse-are scarce in earnings calls and congressional hearings. Worries about runaway super-intelligence have been replaced by worries about “losing” to Beijing.
Key Takeaways
- Every major U.S. AI lab now courts defense contracts, ending a two-year moratorium
- The about-face tracks a broader shift from neoliberal globalization to techno-nationalism
- Cloud giants leverage spy-agency deals for both revenue and regulatory cover
- Defense tech startups promise to disrupt legacy contractors-and secure a slice of $14 trillion in Pentagon spending
- Leaders frame AI competition as a zero-sum clash between democratic and authoritarian systems
