Space Exploration Technologies is reportedly lining up forty billion dollars in debt to buy Nvidia semiconductors, a financing package that would rank among the largest ever tied to artificial intelligence infrastructure and underscore how the chipmaker has become the toll collector for the industry's build-out. Apollo Global Management is said to be leading the transaction, split between ten billion dollars of bank loans and thirty billion dollars of investment-grade debt, though SpaceX, Nvidia, and Apollo have not confirmed the deal. Closing is not expected until twenty twenty-seven.
The financing package
The scale of the raise reflects the capital intensity of the AI race. SpaceX burned through eighteen point four billion dollars of capital expenditure in the second quarter alone, and free cash flow for the first half of the year came in at negative twenty-five billion dollars as the company funded AI data centers, Starlink, and Starship development simultaneously. The cash drain mirrors a broader pattern: Amazon, Alphabet, Meta Platforms, and Oracle collectively issued roughly one hundred ninety-four billion dollars of bonds through early July to finance infrastructure. Lindsay Rosner, who runs multi-sector investing at Goldman Sachs Asset Management, estimates the hyperscalers could sell another four hundred twenty billion dollars of debt next year as build-outs remain a priority.
A compute landlord with contracted revenue
SpaceX has assembled a compute-leasing operation that already carries billions in contracted revenue. Anthropic has committed one point two five billion dollars a month for roughly three hundred twenty-five thousand Nvidia graphics processors. Google holds a contract worth nine hundred twenty million dollars monthly for about one hundred ten thousand of the same chips. Reflection AI, a Nvidia-backed startup, has a one hundred fifty million dollar monthly deal for GB300 chips housed in SpaceX's Colossus 2 facility. On the second-quarter call, management said cloud services revenue of six point seven billion dollars had been booked in the first weeks of the third quarter, with capacity coming online over six months. A separate, unnamed hosting client is slated to add one point one billion dollars a month starting in December.
Musk's exclusive bet on Nvidia
Musk told analysts the company will build "exclusively" on Nvidia, calling the Vera Rubin architecture the best AI computer available. The company targets cumulative compute capacity near ten gigawatts by the end of twenty twenty-seven, an infrastructure footprint that would require sustained capital deployment at a pace few private companies can match.
The debt boom behind the chip boom
Nvidia's data center revenue reached eighty-nine billion dollars in the second quarter, up one hundred seventeen percent from a year earlier. Hyperscalers accounted for forty-eight point seven billion dollars of that total. The financing cycle is self-reinforcing: chip sales enable revenue contracts, which justify debt issuance, which funds more chip purchases. SpaceX's potential forty billion dollar raise is the latest evidence that the AI boom is still being funded by borrowed money at a scale that dwarfs previous technology cycles.