SpaceX Orders $52B in Nvidia GB300 GPUs — and Plans to Lease Them Out

SpaceX has placed an order for roughly $52 billion worth of Nvidia GB300 AI servers — nearly 1 million GPUs across approximately 13,000 racks — making it one of the largest single AI infrastructure commitments ever reported. Elon Musk has publicly acknowledged the order, and the story is bigger than the headline number: SpaceX isn't just building compute for itself. It's building a business around leasing that compute to some of the most resource-hungry AI labs in the world.

The Scale of the Order

According to reporting cited by Sawyer Merritt, the order covers Nvidia's GB300 AI servers — the successor to the GB200 NVL rack systems — with shipments expected to begin in Q4 2026 and continue into Q1 2027. Foxconn has reportedly secured part of the manufacturing contract. At roughly $52 billion, this single order dwarfs the annual AI infrastructure budgets of most Fortune 500 companies.

For context, Musk has previously stated that his companies — SpaceX AI and Tesla among them — would continue ordering Nvidia chips at scale. In May 2025, he noted Nvidia's current superiority in the AI chip market while confirming xAI would also continue acquiring chips from both Nvidia and AMD. The GB300 order appears to be the largest concrete expression of that commitment to date.

SpaceX as an AI Infrastructure Landlord

What makes this story more than a procurement announcement is what SpaceX is doing with the capacity it already has. The company has quietly positioned itself as a major GPU cloud provider, signing multi-year leasing agreements with some of the biggest names in AI:

The Anthropic deal alone — $1.25 billion per month — would generate roughly $15 billion annually if sustained. That's a meaningful revenue stream for a company that has historically funded itself through launch contracts and Starlink subscriptions. According to these reports, the GPU leasing business is becoming a significant and fast-growing third pillar of SpaceX's commercial operations.

The Nvidia Relationship Goes Deeper Than Procurement

SpaceX and Nvidia's partnership extends beyond standard server orders. Nvidia has supplied SpaceX with computing platforms specifically engineered for space environments, including the Space-1 Vera Rubin Module and IGX Thor systems. In October 2025, Nvidia CEO Jensen Huang personally delivered the first DGX Spark — a compact AI supercomputer built on Grace Blackwell architecture — to Musk at SpaceX's facilities. The GB300 order fits a pattern of deepening technical collaboration, not just a vendor relationship.

What This Means Beyond SpaceX

Tesla owners following the AI thread here have a direct stake in this story. Musk has consistently framed Tesla's AI ambitions — FSD, Optimus, Dojo — as requiring massive compute investment. The same logic driving SpaceX's $52 billion GPU bet applies to Tesla's own infrastructure roadmap. As SpaceX builds out what amounts to a private AI cloud, the question of how Tesla's compute needs get served — whether through shared Musk-ecosystem infrastructure or Tesla's own Dojo clusters — becomes more pointed.

Shipments are expected to begin in Q4 2026. By early 2027, SpaceX could be operating one of the densest concentrations of AI compute on the planet — and collecting billions per month from the labs that need it most. Whether that compute advantage eventually flows back into SpaceX's own AI products, or remains a pure infrastructure play, is the story worth watching next. For more on the AI and compute developments shaping Tesla's roadmap, see our AI & Robotics coverage.

This report was curated by the BASENOR Editorial Desk from the sources listed above. Read our editorial standards or email editorial@basenor.com to report an error.