Business Insider reported that SpaceX CEO Elon Musk said at the company’s October 20, 2026, third-quarter earnings call that hyperscalers and dozens of smaller companies had asked to rent its computing capacity. He described a plan to divide compute roughly evenly between internal work and third-party rentals, with the balance able to change month to month.
Musk’s account of rental demand
Musk said the requests could amount to several times the capacity SpaceX would rent if it accepted every one. That was his characterization of demand. He did not name the prospective renters.
A flexible allocation and spot-market pricing
Musk said SpaceX aimed to reserve roughly half its compute for internal work, including scaling Grok and supporting Cursor, and rent roughly half to outside customers. He said the allocation could shift each month.
He also said SpaceX had tried to move some inference workloads to other cloud providers so it could use more compute for training. For outside rentals, Musk described an approach based on effectively spot-market rates, meaning prices tied to current market conditions rather than long-term deals. He said rental rates for older Nvidia H100 chips had risen since those chips were new.
Separate Google and Anthropic agreements
Forbes México separately reported that SpaceX entered a cloud-services agreement with Google on June 5, 2026, covering access to about 110,000 Nvidia computing components. The reported schedule calls for monthly payments of $920 million from October 2026 through June 2029, subject to contract terms.
Forbes México also reported a similar SpaceX-Anthropic arrangement announced in May 2026, with monthly payments of $1.25 billion through 2029. These are separate reported agreements from the flexible spot-market rental approach Musk described.
