The company is in talks to supply multibillion dollar data center capacity for AI, a deal that would put launch, satellite communications, and model training under one vendor for the first time.
SpaceX is in talks with the U.S. Department of Defense to supply multibillion-dollar data center capacity for running AI models, according to the Wall Street Journal, citing people familiar with the matter. The deal is not signed, SpaceX and the Pentagon declined to comment, and the talks could still fall through.
SpaceX already flies the military's payloads, runs the Starshield satellite network that supports intelligence and missile-tracking work, and is building out megawatt-scale data centers designed to train and run large AI systems. No other vendor has offered the Defense Department the same combination, and the offer is the reason SpaceX is telling counterparts it can underbid the cloud companies already in the defense supply chain.
The Pentagon asked Congress for roughly $30 billion to modernize its AI supercomputing arsenal in fiscal 2027, per DefenseScoop's reading of the budget request, and the addressable workload includes model training and inference for intelligence agencies, the military services, and the new AI-integrated command-and-control systems the department is standing up. The big three cloud providers, led by AWS, are the default answer; Amazon has said it will invest up to $50 billion in AI and supercomputing capacity for U.S. government customers. The Pentagon is shopping for additional capacity, not consolidating.
SpaceX is positioning itself as a cheaper alternative to the neocloud tier, the specialist AI-cloud providers that have been the Pentagon's main source of new GPU capacity. The company has told counterparts it intends to compete more directly with CoreWeave, a publicly traded AI-cloud provider, and similar vendors, the Globe and Mail reported. The pitch is not that SpaceX is a better landlord; it is that the launch, the satellite network, and the AI compute are already paid for and operational.
Two recent commercial deals serve as proof of concept. In May, Anthropic agreed to use the full capacity of SpaceX's Memphis data center, gaining 300 megawatts of dedicated AI compute in one of the largest single-site capacity commitments on record. In June, SpaceX signed a multi-year deal giving Google access to roughly 110,000 Nvidia chips and related infrastructure, The Motley Fool reported. The Pentagon would be the third anchor customer on a footprint that is being built for hyperscale, not defense.
The Pentagon is not buying the stack; it is buying capacity. If the contract turns out to be SpaceX acting as a generic GPU landlord, the vertical-integration argument collapses and the deal is just another neocloud procurement with a SpaceX logo on it. The talks are also early: the precise dollar envelope is undisclosed, no draft contract has been reported, and there is no on-record Pentagon rationale. The "could still fall through" line in the original reporting is the falsifier to hold against any follow-up.
The next milestone is a draft contract. If one shows up, the question is which slice of the stack the Pentagon is actually buying: the data center, the satellite link, the launch capacity, or all three. Until then, the model is a pitch.