SpaceX has signed a computing power deal worth up to $6.3 billion with Reflection AI, the open-source artificial intelligence startup founded by former Google DeepMind researchers.
Under the agreement, Reflection will pay $150 million per month beginning July 1, 2026, for access to Nvidia GB300 chips at SpaceX’s Colossus 2 data center near Memphis, Tennessee. The contract runs through the end of 2029, although both parties can exit with 90 days’ notice after the first three months.
The deal adds Reflection to a growing list of outside companies renting compute capacity from SpaceX, following earlier agreements with Anthropic, Google, and Cursor. With committed revenues from those outside clients now exceeding $80 billion through 2029, SpaceX’s Colossus infrastructure has rapidly evolved from an internal AI project into one of the largest commercial compute platforms in the U.S..
Who Is Reflection AI?
Reflection AI was founded in 2024 by Misha Laskin and Ioannis Antonoglou, both veterans of Google’s DeepMind lab. Antonoglou was among DeepMind’s founding engineers and contributed to landmark projects including AlphaGo, AlphaZero, and MuZero.
The startup, currently valued at $25 billion, has not yet released a public frontier model, but it has been building relationships with government and national security customers. In May 2026, Reflection signed a memorandum of understanding (MoU) to support the U.S. Department of Energy’s Genesis Mission, an initiative to build foundational AI infrastructure across 17 national laboratories.
Nvidia has also invested $800 million in Reflection, which creates a notable loop in this deal. Reflection will now train its models on Nvidia hardware purchased and operated by SpaceX, making Nvidia simultaneously an investor in and an indirect supplier to the same customer.
What Reflection Gets From This Deal
For Reflection, the agreement secures access to Nvidia GB300 chips at a time when top-tier GPU capacity remains one of the tightest constraints in the AI industry. The GB300 system combines 72 Blackwell Ultra GPUs with 36 Grace CPUs in a rack-scale setup designed for large-scale AI training and reasoning workloads. Getting reliable access to that kind of hardware at scale is not easy for even well-funded AI startups.
Reflection’s pitch has also found new traction recently. After Anthropic restricted access to its Fable and Mythos models as demanded by the U.S. government, the episode reignited debate about the risks of building critical workloads entirely on closed-model platforms. Reflection is positioning its open-source approach as a direct answer to those concerns.
“Recent events highlight how important open-source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models,” a Reflection spokesperson said. “Our deal with SpaceXAI signals Reflection’s strategic importance within the frontier AI ecosystem, and more compute means more runway to build the world’s best open models at scale.”
Some investors have also drawn comparisons between Reflection and China’s DeepSeek, given both companies’ focus on building competitive open models with efficient resource use.
How SpaceX Became a Compute Landlord
The Colossus complex was originally built to train Grok, the AI chatbot developed within Elon Musk’s technology ecosystem. The facility was never meant to be a commercial compute platform, but that changed as SpaceX began opening Colossus to outside customers.
Anthropic signed on in May 2026 to rent all compute capacity at the original Colossus 1 site, covering more than 220,000 Nvidia GPUs and 300 megawatts of power, in a deal valued at roughly $45 billion through mid-2029. Google followed in early June with a commitment to rent capacity at Colossus 2, paying $920 million per month in a deal worth approximately $30 billion through the same period. SpaceX is also in the process of acquiring Cursor, the AI coding platform, which has been using Colossus infrastructure since April. Reflection now rounds out the customer list as the fourth major outside tenant.
Why This Matters Beyond the Deal Itself
For a company that raised $85.7 billion in its IPO earlier this month on the strength of its rocket and Starlink businesses, the Colossus compute strategy represents a distinct and fast-growing revenue line. Investors have been watching whether SpaceX could expand beyond its traditional businesses, and the accumulation of long-term compute contracts provides a credible answer to that question.
Access to advanced Nvidia chips has become a genuine strategic asset in the current AI environment, and companies that control large-scale GPU infrastructure, along with the data center space and power capacity to run it, are shaping the competitive dynamics of the AI race just as much as the labs building the models. And SpaceX has positioned itself squarely in that category. We only have to watch if SpaceX succeeds in its post-IPO chapter.
