Reflection AI raised $2B and locked in $6.3B of SpaceX compute. It hasn't shipped a model. The Pentagon cleared it anyway.

Reflection AI just raised $2 billion and locked in $6.3 billion of SpaceX compute to build the open-source frontier model that doesn't exist yet.

The company hasn't shipped a single public model.

But it has Pentagon IL6/IL7 clearance, Nvidia's $800 million investment, and GB300 chips waiting in Memphis starting July 1.

This is the most aggressive bet on open-weight AI in history.

Founded in 2024 by two former DeepMind researchers who built AlphaGo, Reflection is targeting the exact gap no one else can fill: frontier performance, open weights, American-made, government-cleared.

Meta's Llama 4 is powerful but not frontier-class. DeepSeek V4 comes close but is disqualified for US government work by origin. OpenAI and Anthropic are closed by design.

Reflection is positioning as the third option.

The $6.3 billion SpaceX deal is the signal. You don't lock in GB300 access for three years unless you plan to use it. The model bottleneck is no longer compute. It's data-pipeline engineering.

For enterprises: if Reflection delivers, your vendor lock-in math changes permanently. Open weights mean you can inspect, fine-tune, and deploy without depending on a single provider's API pricing.

Audit your model procurement strategy now. The closed-lab monopoly just got its first credible challenger with government backing.