An open-source AI startup just committed to one of the largest compute contracts in the industry.
Reflection AI will begin paying SpaceX $150 million per month on July 1, 2026, under a deal running through 2029. The contract gives Reflection immediate access to Nvidia's latest GB300 AI chips and supporting hardware hosted at Colossus 2, SpaceX's data center outside Memphis, Tennessee. Over the full term, the arrangement could exceed $5 billion in total compute spend.
The scale of this deal says something about where the compute arms race stands: even well-funded open-source labs now need to lock in hardware years in advance to stay competitive. SpaceX, meanwhile, is quietly turning its Colossus infrastructure into a revenue-generating cloud business rather than a captive resource for internal AI work.
For a lab pitching open-source as its differentiator, a multi-year, nine-figure-a-month infrastructure bill sets a high bar — one that will require either a very large model or a very patient set of investors to justify.