Deal Overview
According to The Verge (May 1), the US Department of Defense has reached classified-level cooperation agreements with three AI giants:
| Company | Cooperation Area | Previous DoD Relationship |
|---|---|---|
| OpenAI | Large model inference & decision support | Already has Maven Smart System contract |
| Gemini model + cloud infrastructure | Long-term Android military version supplier | |
| Nvidia | GPU compute + AI training platform | Primary military GPU supplier |
| Anthropic | Not participating in this round | Previously the main model supplier for classified info |
Key Reasons for Anthropic’s Exclusion
This isn’t a question of technical capability, but a shift in procurement logic:
1. Scale Over Safety Narrative
Anthropic’s core selling point is “safety-first” AI development, but in defense procurement:
- OpenAI’s GPT series has been battle-tested at scale
- Google’s cloud infrastructure covers the DoD’s global network
- Nvidia’s GPU compute is the foundational layer for training and inference
Together, the three form a complete chain “from chip to application.” Anthropic only has the model layer.
2. Capacity Bottlenecks
Anthropic’s inference infrastructure scale is far smaller than its three competitors. The DoD needs:
- Global deployment
- High-concurrency inference
- Offline/edge computing scenarios
Anthropic currently lacks the infrastructure to support these scenarios.
Investment Logic
| Target | Military AI Exposure | Growth Space | Risk |
|---|---|---|---|
| Microsoft (OpenAI) | High: Maven contract + Azure Gov | DoD AI budget continues growing | Public opinion pressure |
| Medium-High: Project Maven history + Gemini | Cloud infrastructure replacement cycle | Internal employee opposition | |
| Nvidia | High: GPU is the only choice | Training + inference dual demand | Export control risk |
| Anthropic | Zero this round | May pivot to other government agencies | Marginalization risk |
Industry Impact
- AI Military-Industrial Complex taking shape: From chips (Nvidia) to cloud (Google) to models (OpenAI), defense AI procurement is forming a closed loop
- Anthropic’s “safety premium” fails in government markets: Safety ethics is a selling point in commercial markets but may be read as “capability limited” in military markets
- Chinese military AI acceleration window: Rapid US defense AI integration will stimulate equivalent investments from peer nations