SpaceX is now pulling in roughly $2.32 billion per month by renting GPUs to three of the most aggressive AI companies on the planet. That’s an annualized revenue run rate of approximately $28 billion, a number that would make most publicly traded cloud companies deeply uncomfortable.

The three customers: Anthropic, Google, and Reflection AI. The latest deal, finalized around June 22, 2026, locks Reflection AI into paying SpaceX $150 million per month for access to advanced GB300 GPUs, with the contract stretching through 2029 and totaling a potential $6.3 billion over its lifespan. It includes flexible exit clauses, but the sheer scale of the commitment tells you everything about where AI compute demand is heading.

From rockets to racks: how SpaceX became an AI landlord

SpaceX’s Colossus data centers were originally built to support xAI, Elon Musk’s AI venture. But excess capacity at those facilities created an opportunity that was too lucrative to ignore. Rather than let expensive GPU clusters sit underutilized, SpaceX pivoted the infrastructure into a high-margin compute leasing business. Pricing for the premier GB300 GPUs likely exceeds $10 per hour, placing SpaceX squarely in the “neocloud” category, a term for newer entrants offering specialized AI compute that competes directly with legacy cloud providers like AWS, Azure, and Google Cloud Platform.

For context, CoreWeave, perhaps the best-known neocloud provider, went public and currently carries a valuation of around $60 billion. SpaceX’s AI compute revenue run rate already surpasses CoreWeave’s scale, and SpaceX hasn’t even held its IPO yet.

The IPO math gets interesting

These contracts were signed during mid-2026, a period when SpaceX has been actively preparing for an Initial Public Offering targeting a valuation north of $1.7 trillion. SpaceX already generates income from Starlink, its satellite internet constellation, and from government and commercial launch contracts. Adding a $28 billion annualized AI compute business on top creates a diversification story that most pre-IPO companies could only dream about.

What this means for investors and the broader AI market

The emergence of decentralized compute networks like Akash, Render, and io.net has been partly driven by the thesis that centralized cloud providers would eventually face capacity constraints. SpaceX’s ability to sign $28 billion in annualized contracts specifically because established providers couldn’t meet demand validates that thesis, even if the solution in this case is another centralized player rather than a decentralized protocol.

Reflection AI’s contract includes flexible exit clauses, meaning the $6.3 billion total isn’t guaranteed. GPU technology evolves rapidly, and today’s cutting-edge hardware becomes tomorrow’s commodity. There’s also concentration risk: three customers generating $28 billion is impressive but fragile, and losing any single contract would represent a massive revenue hit. SpaceX remains a private company, meaning retail investors can’t directly participate until the IPO actually happens.