SpaceX is in talks for $40 billion in financing earmarked to purchase chips from Nvidia, according to sources cited by Yahoo Finance, in a transaction that would rank among the largest single-vendor AI hardware commitments reported to date.

The figure, attributed to unnamed people familiar with the matter, points to a procurement scale normally associated with hyperscale cloud operators rather than a launch services provider. SpaceX has not publicly disclosed a specific Nvidia product family, node, or wafer-size commitment tied to the proposed financing, and the company declined to comment on the report. Nvidia likewise has not confirmed the deal.

What financing structure is on the table?

Details on the debt-versus-equity mix, tenor, and lead arrangers remain undisclosed. The $40 billion total would, at the high end, rival some of the largest corporate debt facilities arranged in the technology sector over the past two years. Whether the package includes export-controlled parts subject to U.S. licensing rules is also not addressed in the available reporting.

Why would a launch company buy AI silicon?

SpaceX operates the Starlink broadband constellation, which carries an onboard compute and beam-forming workload across more than 7,000 satellites in low-Earth orbit, alongside the company’s ground infrastructure and internal machine-learning pipelines. Nvidia’s data-center accelerators — sold under the H100, H200, and Blackwell families, fabricated on TSMC’s 4NP, 4N, and a custom 4NP-derived process node respectively — are the dominant discrete GPUs in the market. None of the reporting reviewed ties the SpaceX purchase request to a specific SKU or generation.

What does the deal signal about Nvidia demand?

Even unverified, the headline figure reinforces the picture Nvidia has painted in consecutive quarters: customer demand running ahead of supply, and allocations effectively rationed. CEO Jensen Huang told analysts on Nvidia’s most recent earnings call that the company expects Blackwell production to scale through calendar 2025, with capacity sold out multiple quarters out. A $40 billion order, if executed, would consume the output of a meaningful share of an advanced packaging line for a sustained window, depending on whether it is staged over years or front-loaded.

Where does the supply chain sit?

Any Nvidia accelerator ships from TSMC’s CoWoS-L advanced packaging lines, currently ramping at sites in Taiwan and, more recently, at JASM in Kumamoto, Japan, and through Amkor’s facility in Arizona. Allocation of those CoWoS slots — not raw wafer starts — is the binding constraint industry analysts have flagged repeatedly. A reported SpaceX commitment would compete for the same CoWoS throughput that cloud providers, sovereign AI initiatives, and enterprise system integrators are already queued for.

What is confirmed versus estimated?

The only confirmed data point in the available reporting is that SpaceX is seeking roughly $40 billion in financing intended for Nvidia chip purchases. Product mix, delivery schedule, financing counterparties, and any supply-agreement terms are not part of the public record. Nvidia and SpaceX have not issued statements verifying or rejecting the framing of the deal.

What to watch next

The size of any announced facility, the identity of the lead arrangers, and the Nvidia product family specified in the final term sheet will determine whether this becomes the largest single-customer chip purchase on record or a more conventional multi-year procurement wrapped in a large headline figure.