HAWTHORNE, Calif. — The headlines this week centered on Anthropic, but the more durable story is the customer book forming behind it. SpaceX's Colossus supercomputer complex has quietly become home to a roster of blue-chip artificial-intelligence tenants, positioning the company as a neutral landlord for an industry that cannot build capacity fast enough.
Anchoring that book is Anthropic, which under SpaceX's spring IPO filings committed to roughly $1.25 billion a month — close to $45 billion through May 2029 — for dedicated capacity. But it is no longer alone. Comparable compute agreements with Google and Reflection AI reportedly carry similar structures, giving SpaceX a diversified set of marquee customers rather than a single anchor tenant.
The landlord model
What makes the arrangement powerful is its neutrality. SpaceX does not need to win the model wars to win the compute wars; it simply needs to own the scarce resource every lab is chasing. That is a page from the playbook Musk's companies have run before, and analysts have increasingly built the bull case around this compute economics.
A diversified tenant list also changes the risk math. Each of the major contracts reportedly includes 90-day exit rights for both sides, which sounds fragile until you count the tenants. With several blue-chip customers under roof, no single departure is decisive — and in a tight market, any opening is quickly backfilled.
