SpaceX is considering raising $40 billion to finance the purchase of advanced graphics processing units and related systems, according to a report, a move that would push the privately held rocket company deep into the capital-intensive race to build artificial intelligence infrastructure.

The reported figure would rank among the largest debt raises ever attempted by a private company and would be used largely to buy Nvidia chips, the hardware that dominates AI model training and inference. SpaceX has not confirmed the plan, and the report did not specify a timeline or identify lenders.

The gamble carries obvious risk: Nvidia's most advanced chips sell for tens of thousands of dollars each, and a $40 billion order implies a purchase on a scale few buyers outside Microsoft, Meta, Google and Amazon have attempted. Servers, networking gear, power and cooling to run those chips can add billions more.

SpaceX's core business is launching satellites and rockets, not operating data centers. The company runs Starlink, its satellite internet service, which has more than 5 million subscribers worldwide, and has said it uses machine learning to manage network traffic and autonomous collision avoidance for its spacecraft. AI compute would give it control over that infrastructure rather than renting it from cloud providers.

Taking on debt to buy depreciating hardware marks a shift for a company that has historically funded itself through equity raises, government contracts and launch revenue. SpaceX was valued at roughly $350 billion in a December 2024 tender offer, according to reports at the time, and CEO Elon Musk has said the company's Starship program alone consumes billions of dollars annually.

Nvidia's data center revenue reached $35.1 billion in its fiscal third quarter, reported in November 2024, a 112% increase from a year earlier. Demand has outstripped supply for successive generations of its chips, and buyers have resorted to multiyear prepayments and direct negotiations with CEO Jensen Huang to secure allocations.

Musk has publicly complained about constraints on AI hardware. His separate company xAI, which builds the Grok chatbot, assembled a cluster of 100,000 Nvidia H100 chips in Memphis, Tennessee, in 2024, a buildout he described as the fastest of its kind. xAI has raised billions in equity and debt to fund that expansion.

Whether lenders will accept data center hardware as collateral for a $40 billion loan is unsettled. GPUs depreciate faster than real estate or pipelines, and chip generations turn over roughly every 18 to 24 months. Lenders weighing such a deal would be underwriting an asset whose resale value could fall sharply if AI spending slows.

SpaceX did not respond to a request for comment on the reported financing plan.