SpaceX Leases $6.3 Billion of Nvidia Compute to an Open-Weight Lab, Turning Colossus Into a Merchant Cloud

SpaceX has agreed to lease Nvidia GB300 capacity inside its Colossus 2 data center near Memphis to Reflection AI, an open-weight model startup, for about $150 million a month beginning in July, with payments reaching roughly $6.3 billion if the contract runs through 2029. Either side can exit with 90 days' notice after the first three months.

The arrangement creates a circular flow of money. Nvidia invested $800 million in Reflection, and Reflection is now paying to use Nvidia accelerators that SpaceX bought. Reflection gets the reasoning-class GB300 graphics processing units (GPUs) it needs to train models without building its own facility, SpaceX earns money from idle capacity, and Nvidia profits from its chips a second time. The deal follows earlier Colossus tenancy agreements with Anthropic, Google and Cursor, confirming SpaceX's move into commercial compute leasing.

The economics are straightforward. A monthly rent of $150 million implies an annual rate near $1.8 billion for a single tenant, a figure that only works if Reflection can turn that compute into models that earn back the cost. For an open-weight lab that gives its weights away, revenue must come from services, enterprise support and inference rather than licensed access. That is the assumption now being financed at a scale of several billion dollars.

What is verified is the structure of the deal and the dollar figures, which multiple outlets report consistently. What remains a claim is the strategic logic, namely whether leased capacity at this price is a step toward durable demand or a commitment that will look expensive if model revenue lags.