SpaceX, the company best known for launching rockets and beaming internet from orbit, is now pulling in some of the largest AI infrastructure contracts on the planet. Google has agreed to pay $11 billion annually to lease roughly 110,000 Nvidia GPUs from SpaceX’s data centers, a deal starting October 2026 and running through June 2029 that could exceed $30 billion in total value.

Meanwhile, the cost of building out AI infrastructure has climbed more than 50% in just two weeks, jumping from approximately $50 billion to $75 billion per gigawatt.

The deals reshaping AI compute

Google’s arrangement with SpaceX works out to roughly $920 million per month.

Anthropic, the AI safety company behind the Claude model, has contracted SpaceX to lease AI compute capacity at $1.25 billion per month. That annualizes to $15 billion, making Anthropic’s deal even larger than Google’s on a yearly basis.

Combined, these two contracts alone represent $26 billion in annualized revenue for SpaceX’s AI infrastructure business.

The Google contract was disclosed through an amendment to SpaceX’s S-1 filing, signaling that the company views AI compute leasing as a material revenue stream worth highlighting to potential public market investors.

Why compute costs are exploding

The jump from $50 billion to $75 billion per gigawatt in infrastructure costs captures the full picture. It’s not just the hardware. It’s the power delivery, the cooling infrastructure, the real estate, and the engineering talent required to stitch it all together.

The 110,000 Nvidia GPUs that Google will access through SpaceX represent a substantial cluster. For context, training frontier AI models requires tens of thousands of GPUs working in concert.

SpaceX’s quiet transformation

SpaceX has historically generated revenue from two primary sources: launch services and Starlink subscriptions. The AI leasing business represents a fundamentally different revenue profile, one characterized by long-term contracts with predictable, recurring cash flows.

SpaceX is preparing for a potential public offering, and an S-1 filing that shows $26 billion in annualized AI compute contracts from Google and Anthropic tells public market investors something important: this isn’t just a space company anymore.

What this means for investors

The 50% surge in compute infrastructure costs is a leading indicator for the entire AI supply chain. Companies that build data centers, manufacture cooling systems, produce power equipment, and Nvidia itself all stand to benefit from a market where demand is repricing infrastructure this aggressively.

When two companies, Google and Anthropic, account for $26 billion in annualized revenue from a single provider, any disruption to those relationships would be significant. Investors evaluating SpaceX’s eventual public offering will need to weigh the attractiveness of massive recurring revenue against the risk of customer concentration.

Anthropic’s willingness to pay $1.25 billion per month for compute access, annualizing to $15 billion, likely represents a substantial portion of Anthropic’s total operating budget, suggesting that the company views securing compute capacity as existential, not optional.