Elon Musk’s SpaceX has inked another major computing power agreement, this time with Reflection AI, an open-source artificial intelligence startup founded by former Google DeepMind researchers. The multibillion-dollar deal underscores how rapidly compute capacity has become the most strategically valuable commodity in the global AI race, with advanced Nvidia chips now functioning as the ultimate leverage point for companies training frontier models.
Under the agreement, Reflection will secure immediate access to Nvidia’s top-tier GB300 chips housed at SpaceX’s Colossus 2 data center in Memphis, Tennessee. The startup will pay SpaceX $150 million per month beginning July 1, 2026, extending through 2029. If the contract runs its full term, the total commitment reaches approximately $6.3 billion, though both companies retain the right to terminate with 90 days’ notice after an initial three-month minimum commitment period. The structured approach gives Reflection breathing room while securing SpaceX’s revenue predictability.
This arrangement marks another decisive move in SpaceX’s transformation of Colossus from a proprietary infrastructure project into a commercial computing platform generating revenue streams from major AI players. Colossus was originally constructed to power Grok, Musk’s AI chatbot intended to rival OpenAI’s ChatGPT. SpaceX has already struck similar compute lease agreements with Google, Anthropic, and Cursor, which SpaceX is now acquiring in an all-stock transaction. The pattern is clear: SpaceX has discovered that leasing scarce computational resources may ultimately prove more lucrative than rocket launches alone.
READ ALSO: Allbirds New CEO Steps In with Big AI Ambitions but Zero Staff.
What distinguishes the Reflection deal is the strategic positioning around open-source AI development. Reflection, backed by Nvidia with a $25 billion valuation and $800 million in funding from the chip manufacturer, has not yet released a public frontier-scale open-source model. The company has instead cultivated deep relationships with US government agencies and national security customers, working directly with the Department of Energy’s Genesis Mission and participating in broader Pentagon AI initiatives. This government alignment gives Reflection distinct advantages in an era when policymakers increasingly recognize the risks inherent in depending entirely on closed-model providers.
The timing of this deal capitalizes on growing momentum in open-source AI. Anthropic’s recent decision to restrict access to its most powerful models after the Trump administration threatened to block foreign nationals from using them has triggered serious conversations about concentrating too much intelligence capability in the hands of any single private company. A Reflection spokesperson noted that recent developments highlight how important open-source systems have become, with governments and enterprises now recognizing the costs and dangers associated with exclusive dependence on closed AI solutions.
For SpaceX, the agreement reinforces a broader strategic calculation: access to advanced Nvidia hardware remains the primary constraint preventing most companies from building competitive large language models. By positioning itself as a reliable supplier of premium GPU capacity, SpaceX has essentially become an AI infrastructure company disguised as an aerospace firm. Whether this pivot represents deliberate strategy or opportunistic response to surging enterprise demand remains open to interpretation, but the financial numbers speak for themselves. With Google reportedly committing to around $30 billion in compute costs and Anthropic committed to approximately $45 billion through 2029, SpaceX’s Colossus infrastructure is quietly becoming one of the most economically important assets in the entire Musk portfolio.
