SpaceX Signs $6.3 Billion AI Compute Deal With Reflection — Its Rockets Are Now a Side Business

SpaceX has signed a computing power agreement worth up to $6.3 billion with Reflection AI, an open-source artificial intelligence startup, marking one of the clearest signals yet that Elon Musk's rocket company has transformed itself into a genuine AI infrastructure provider. The deal, reported Monday, June 22, 2026, by CNBC and confirmed by Bloomberg and multiple other outlets, will see Reflection pay SpaceX $150 million per month starting July 1, 2026, with the agreement structured to run through the end of 2029 if both parties see it through to term. What the Deal Actually Provides Under the agreement, Reflection will gain immediate access to Nvidia GB300 chips — among the most advanced AI processors currently available, used for training and running frontier-scale AI models — housed at SpaceX's Colossus 2 data center in Memphis, Tennessee. The arrangement includes a built-in flexibility clause: either company can terminate the deal with 90 days' notice, but only after the first three months have elapsed, meaning both sides have effectively committed to at least one full quarter of payments before that exit option becomes available. If the deal runs its full course through 2029, the total value would reach approximately $6.3 billion — a substantial sum for a three-year-old AI startup, and a clear indicator of just how capital-intensive frontier AI development has become across the industry. Reflection AI: The Company Behind the Deal Reflection AI was founded in 2024 by Misha Laskin and Ioannis Antonoglou, both alumni of Google's DeepMind research lab. The startup's core mission is building open-source AI models and automation tools, with a particular focus on coding applications — positioning itself as an alternative to the closed, proprietary model approach favored by companies like OpenAI, Anthropic, and Google itself. The company has not yet released a public frontier open-source model, but it has built considerable momentum on a different front: government and national security relationships. Reflection is involved with the US Department of Energy's Genesis Mission and has received clearance to operate within Pentagon classified AI networks — credentials that speak directly to the strategic case the company has been making for its existence. That strategic case got an unexpected boost from recent events. A Reflection spokesperson explicitly tied the deal's importance to broader anxieties about AI infrastructure dependency, stating: "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." The comment appears to reference the recent episode in which the US government ordered Anthropic to suspend access to its most advanced AI models for foreign nationals worldwide — an event that, according to reporting, intensified debate across the industry about the vulnerabilities inherent in relying on closed, single-vendor AI systems that can be abruptly restricted. Reflection has previously raised approximately $800 million in funding, with Nvidia among its most notable backers, pushing the startup's valuation to as high as $25 billion. An Unusual Loop: Nvidia Investor, Supplier, and Beneficiary All at Once One of the more curious dimensions of the deal is the layered relationship it creates between Nvidia, Reflection, and SpaceX. Nvidia has invested roughly $800 million directly into Reflection. Reflection will now run much of its compute workload on Nvidia GB300 chips that SpaceX has separately acquired and houses in its own data center. The arrangement effectively makes Nvidia simultaneously an investor in Reflection, an indirect supplier to the same company through SpaceX's infrastructure, and a financial beneficiary of the very compute deal Reflection is now paying for — a closed loop that illustrates how deeply intertwined ownership stakes, hardware supply, and infrastructure financing have become across the current AI boom. How SpaceX Became an AI Infrastructure Company To understand why a rocket and satellite internet company is now in the business of leasing AI compute to outside startups, it helps to look at the origin of the infrastructure itself. SpaceX's Colossus 2 data center facility in Memphis was originally built to train and run xAI's Grok chatbot models — xAI being a separate Musk-founded AI company. But rather than limiting the facility's use purely to internal Grok development, SpaceX has increasingly opened up Colossus's spare capacity to outside AI companies as a commercial revenue stream. The Reflection deal is far from SpaceX's first such arrangement. According to reporting, SpaceX has already struck similar compute agreements with Anthropic, Google, and Cursor — the latter an AI coding platform that SpaceX is reportedly now in the process of acquiring outright. Reported figures for some of these prior deals are staggering in their own right: Google's compute arrangement with SpaceX has been valued at roughly $920 million per month, while Anthropic's commitment has been reported at approximately $45 billion over its term, structured as a 180-day lease with a 90-day cancellation notice. Google's broader compute commitment to SpaceX has separately been cited at a total value of roughly $30 billion through approximately mid-2029. Taken together, these deals paint a clear strategic picture: SpaceX is increasingly treating its data center infrastructure as a standalone revenue-generating business line, distinct from — though funded in part by — its core rocket launch and Starlink satellite internet operations. What It Means for SpaceX's Valuation Story The Reflection deal lands at a particularly active moment for SpaceX as a company navigating public capital markets. SpaceX recently completed its initial public offering, and the company disclosed holding more than $100 billion in cash following the listing. Just days before the Reflection announcement, SpaceX also launched its inaugural senior unsecured notes offering — a bond sale that reflects the company's growing need to finance massive infrastructure buildouts of the kind that underpin deals like this one. Notably, SpaceX's stock fell roughly 10% on the same Monday the bond offering was announced, a reminder that even amid a strong run of high-value AI compute deals, the company's stock remains sensitive to the scale of capital it is raising and deploying. The broader implication, as several financial outlets have noted, is that SpaceX's investment story is increasingly becoming as much about AI infrastructure as it is about aerospace. Whether that reflects a deliberate long-term strategic pivot by Musk, or simply an opportunistic response to overwhelming enterprise demand for scarce GPU compute capacity, remains an open question — but the scale and frequency of these deals suggest the company is leaning into the AI infrastructure business with real conviction. A Benchmark Price for the Industry Beyond its significance for SpaceX and Reflection specifically, the deal's pricing structure offers a useful real-world data point for the broader AI infrastructure market. At $150 million per month for access to a defined slice of Nvidia GB300 capacity, the agreement establishes a concrete benchmark for what large-scale, frontier-tier GPU access actually costs in 2026's market conditions — information that newer, decentralized compute marketplaces such as Render, Akash, and io.net, which have long argued they can undercut centralized providers on price, will now have a very specific figure to measure themselves against. What Happens Next For Reflection, the deal provides what its own spokesperson described simply: "More compute means more runway to build the world's best open models at scale" — capacity the company says it needs to compete with closed frontier labs while still pursuing its national security and government client relationships. For SpaceX, the agreement adds another multibillion-dollar revenue commitment to a rapidly growing AI infrastructure portfolio that increasingly rivals its traditional aerospace business in scale and strategic importance. And for the AI industry more broadly, the deal adds to a now-familiar pattern in 2026: compute capacity itself has become one of the most valuable and fiercely contested commodities in technology, with companies that control data center infrastructure — whether built originally for rockets, search engines, or chatbots — increasingly finding themselves in the business of selling that capacity to whoever needs it most.