SpaceX grants Reflection, the open-source startup, priority access to Nvidia “Blackwell Ultra” chips inside its Colossus 2 data centre, cementing a secure American alternative to foreign AI models.
On 22 June 2026, SpaceX announced a massive compute-capacity lease with Reflection AI, an open-source AI startup. The contract is worth up to $6.3 billion through to the end of 2029. Starting 1 July 2026, Reflection AI will pay SpaceX a fixed premium subscription fee of $150 million per month. The contract features a flexible “hybrid” exit clause allowing either party to terminate the agreement with a 90-day notice period after the initial three-month trial.
Reflection AI is currently valued at $25 billion and is backed by a massive $800 million strategic investment directly from Nvidia, explaining how a startup can afford a $150 million monthly compute bill. This marks SpaceX’s fourth major compute customer for its Memphis-based infrastructure. Alongside leasing space to incumbents like Google and Anthropic, SpaceX also recently signed an infrastructure deal with—and is in the process of acquiring—the popular AI coding tool company Cursor in an all-stock transaction valued at $60 billion.
The lease grants Reflection AI immediate, priority access to Nvidia’s latest cutting-edge GB300 “Blackwell Ultra” chips. These accelerators feature advanced liquid-to-chip cooling and fifth-generation NVLink connections to smoothly train trillion-parameter models. The hardware will run out of SpaceX’s newly built Colossus 2 data centre complex located near Memphis, Tennessee.
A Reflection AI spokesperson stated that the massive influx of compute gives them the “runway to build the world’s best open models at scale.” The company is strategically positioning its upcoming open-weight architectures as a highly secure, American open-source alternative to China’s DeepSeek. Additionally, they have established formal deep ties with the U.S. government, working directly on the Department of Energy’s Genesis Mission and various Pentagon AI initiatives.
