How the US Military Turned AI Into a Full‑Stack War Engine

1. From ethics refusal to AI‑first combat in three months

Three months before 1 May 2026 the companies publicly said they would not supply AI for warfare. On that date they signed contracts with the U.S. Department of Defense to operate AI for the armed forces, aiming to create an “AI‑first fighting force” where AI makes battlefield decisions.

“These agreements accelerate the transformation toward establishing the United States military as an AI‑first fighting force and will strengthen our warfighters’ ability to maintain decision superiority across all domains of warfare.”

The Pentagon’s architecture is a four‑layer closed loop: data synthesis, battlefield situational awareness, combat decision making, and action execution, making AI the central brain of warfare.

2. Full‑stack AI system supplied by eight firms

NVIDIA – provides GPU compute for training war‑AI models.

OpenAI + Google – supply the most capable AI models that serve as the combat‑decision brain.

SpaceX – delivers continuous data links via Starlink, ensuring uninterrupted battlefield connectivity.

Microsoft + AWS + Oracle – operate secure cloud infrastructure for running the closed‑loop system.

Reflection AI – developing open‑source models intended to compete with DeepSeek.

The contracts bypass a “single‑point application” stage and implement the full‑stack closed loop, allowing AI to identify targets and issue attacks within a single instant. The agreements also grant AI access to the classified IL‑6 (confidential) and IL‑7 (top‑secret) networks, enabling processing of the most sensitive military information and participation in high‑level decisions.

3. Anthropic excluded as a supply‑chain risk

Anthropic was placed on the Pentagon’s supply‑chain risk list because it refused to include an “any lawful use” clause, citing concerns that such a clause could enable mass surveillance or fully autonomous lethal weapons. Consequently the Pentagon signed with Anthropic’s competitors, even though Anthropic recently released the cybersecurity‑focused model Mythos, which can detect software vulnerabilities.

4. Funding and operational reality

The contracts allocate $54 billion to develop autonomous weapons, a sum sufficient to acquire dozens of unicorn‑scale technology firms. The operational loop now includes AI‑driven target identification, optimal strike planning, and direct control of drones and missiles, with human operators limited to confirmation or moral responsibility.

5. Technical gaps compared with the U.S. effort

Front‑line AI models – U.S. leverages GPT‑4o, Claude and Gemini; domestic models still lag.

Satellite communications – Starlink is deployed in combat; China’s satellite internet is catching up.

Classified network security – IL‑6/IL‑7 closed‑loop is operational; other militaries are still building AI‑enabled systems.

Unmanned combat systems – Integrated ISR‑strike drones are in use; system maturity is improving elsewhere.

AI‑chip compute – NVIDIA H100/H200 GPUs are in mass production; domestic GPU development continues.

6. The AI‑war era is now

The Pentagon’s full‑stack loop is live, $54 billion has been spent, and the eight tech giants are deeply integrated, making autonomous AI control of warfare a present reality.

7. Closing note

The article raises questions about what the companies’ models will enable, who they will target, and how to prevent loss of control, urging security professionals to strengthen their defenses.