SpaceX signs $1 billion-a-month AI hosting deal

SpaceX has signed an artificial intelligence hosting agreement worth about $1.11 billion a month, giving the company greater confidence that it can reach a $100 billion annual recurring revenue run rate by the end of the year.

Chief Financial Officer Bret Johnsen disclosed the agreement Thursday during a fireside chat with Goldman Sachs analyst Eric Sheridan at the Goldman Sachs Communacopia and Technology Conference.

SpaceX signed the hosting agreement earlier this month, Johnsen said. The contract is scheduled to begin December 1 and generate about $1.11 billion a month. That's about $13 billion in annual recurring revenue, or ARR, based on a full year at the initial monthly rate.. It doesn't mean SpaceX will collect the entire amount this year.

The agreement is the latest sign of demand for AI computing capacity. It also provides SpaceX with a source of cash as the company invests in terrestrial data centers, orbital computing, Starship, Starlink broadband and a future satellite-based mobile service.

For investors, the near-term opportunity is selling access to scarce AI infrastructure. The longer-term bet is that SpaceX can use reusable rockets to move a meaningful portion of that infrastructure into orbit.

Deal supports $100 billion revenue target

Johnsen said the latest hosting agreement increased management’s confidence in reaching a $100 billion ARR run rate by year-end.

ARR is commonly used by subscription businesses to express the annual value of recurring contracts. In this case, SpaceX would be annualizing the revenue it expects to generate in December.

“We have even more conviction now about that $100 billion ARR target by the end of this year,” Johnsen said.

Investors should distinguish that run rate from revenue recognized during the year. A company exiting December at an annualized rate of $100 billion hasn't necessarily generated $100 billion during the preceding 12 months.

The length of SpaceX’s AI hosting contracts also matters. Johnsen said most have involved an initial commitment of about 90 days with an additional 90-day period, amounting to roughly six months.

Short agreements give SpaceX the flexibility to reclaim computing capacity for its own AI products. They also provide less revenue visibility than multiyear contracts would.

Management is willing to accept that tradeoff because it doesn't want to promise its computing resources to outside customers indefinitely.

“You don’t want to get in a situation where you constrain your own products in the years to come because you were giving away the compute forever,” Johnsen said.

SpaceX builds a terrestrial AI business

SpaceX expects to finish this year with slightly more than two gigawatts of terrestrial computing capacity. It is targeting the deployment of five to 10 gigawatts next year.

A gigawatt measures electrical power. In the AI industry, power capacity has become a convenient way to describe the potential size of a data-center operation because AI processors consume immense amounts of electricity.

Johnsen said SpaceX would not have issued that projection without visibility into the necessary power supplies, facilities and government permits.

Power is increasingly becoming the limiting factor for AI development. Advanced processors are scarce, but the industry is also struggling to find electrical grids that can support enormous clusters of chips.

SpaceX says its experience building large and complex infrastructure gives it an advantage. The company serves as its own general contractor on some projects, installing power and constructing facilities instead of relying entirely on outside developers.

The one critical component it doesn't control is the supply of AI processors.

Johnsen said SpaceX is exclusively using Nvidia chips and expects its relationship with the semiconductor company to help it obtain the necessary allocations. Nvidia allocation refers to the number of processors a customer is allowed to purchase when demand exceeds the available supply.

AI investments may repay themselves quickly

SpaceX expects its computing capacity to generate approximately $30 to $50 in revenue per watt next year, with current deals near the upper end of that range, Johnsen said.

Revenue per watt measures how effectively an AI operator turns electricity and computing capacity into sales. The metric allows investors to compare the earning power of data centers of different sizes.

Johnsen said the company’s investments in AI infrastructure could have payback periods of less than one year. A payback period measures how long an asset takes to generate enough cash to cover its original cost.

That would make spending on AI processors different from traditional aerospace capital expenditures.

“Not all capital is the same,” Johnsen said.

A graphics processing unit, or GPU, can begin generating hosting revenue shortly after it's installed. A launch tower or hangar may produce returns over several decades.

Johnsen also noted that older generations of Nvidia processors continue to command meaningful prices. That residual value reduces the risk of buying expensive hardware and may make lenders more willing to finance the equipment.

The result could be a new asset class built around GPU financing. Banks and other lenders could provide capital secured by AI processors much as they finance aircraft, vehicles or industrial equipment.

Hosting revenue can finance proprietary AI products

SpaceX is currently renting computing capacity to other AI developers while expanding its own products.

Johnsen cited relationships with Anthropic and Google and said demand for available infrastructure is widespread. The company’s customers appear willing to pay attractive prices because access to power and advanced processors remains severely constrained.

Third-party hosting gives SpaceX an immediate way to monetize its data centers. It may also establish a minimum return for the company’s AI investments while its proprietary products develop.

Management’s eventual goal is to generate more revenue by running its own AI models and applications on the infrastructure.

That creates a strategic balancing act. Renting capacity produces predictable revenue, but reserving it for SpaceX’s own products could provide greater returns if those products become competitive.

Orbital computing is bigger bet

SpaceX ultimately wants to move large amounts of AI computing into the Earth's orbit.

Orbital compute refers to satellites equipped to process AI workloads in space, where the temperature is minus 450 degrees Fahrenheit (-268 °C). Rather than functioning primarily as communications relays, these satellites would operate as data centers powered by free energy from the sun .

Johnsen said SpaceX is targeting the launch of its first orbital-computing satellites next year, followed by significant deployments during 2028.

The company plans to adapt the same underlying spacecraft platform, known as a satellite bus, used for its V3 Starlink satellites. SpaceX would replace the communications payload with computing equipment and add more solar-generating capacity.

Johnsen argued that orbital data centers could avoid several constraints facing facilities on Earth, including electricity shortages, cooling costs, real estate expenses and permitting delays.

Space-based computing comes with its own formidable challenges. Equipment must withstand radiation, temperature changes and the physical stresses of launch. Repairs and hardware replacements are far more difficult. SpaceX must also transmit enormous amounts of data between orbit and Earth.

The economics depend heavily on lowering launch costs through Starship.

“The moment when it becomes better than terrestrial is all about reusability,” Johnsen said.

Starship determines the cost equation

SpaceX designed Starship to be fully reusable, meaning both the booster and upper stage could fly repeatedly instead of being discarded after each mission.

Reusability spreads manufacturing costs across multiple flights and lowers the cost of carrying each kilogram into orbit. SpaceX has already demonstrated that model with the Falcon 9 rocket, whose first-stage boosters have been reflown hundreds of times.

The company is now working toward recovering and reusing both stages of Starship.

Johnsen said a recent flight completed a precision ocean landing of the second stage. SpaceX recovered the vehicle, allowing engineers to inspect its heat shield and learn how the hardware performed during reentry.

The company expects an upcoming Starship flight to carry production V3 Starlink satellites. Unlike earlier demonstration missions, it will be a revenue-generating flight.

“For the CFO, it’s very exciting because it’s going to be a revenue-generating flight,” Johnsen said. “I love the demonstration flights, and now I’m excited that we’re moving into production cadence.”

Production cadence means launching rockets regularly and predictably rather than treating every mission as an experimental event.

Johnsen suggested orbital computing could approach the cost of terrestrial infrastructure as soon as next year. That is an ambitious forecast and depends on Starship achieving frequent flights, successful recovery and economical reuse.

Starlink expands among corporate customers

SpaceX’s connectivity business provides another source of growth while its AI plans develop.

Starlink began as a broadband service for consumers who lacked reliable terrestrial internet access. Johnsen said it now has the reliability needed to serve corporate customers, including service-level agreements that guarantee specified performance standards.

Businesses can use Starlink as their main internet connection or as a backup if terrestrial networks fail.

SpaceX also has a substantial backlog of airline installations and sees opportunities across cruise ships, yachts, trains and other vehicles. These arrangements generate commercial revenue while exposing millions of passengers to the Starlink service.

Johnsen said awareness remains one of the company’s biggest obstacles outside the United States despite Starlink operating in more than 170 countries.

Management also expects demand for satellite connectivity to grow as autonomous vehicles, robots and other AI-powered machines require uninterrupted data connections.

S atellite mobile service targeted for 2028

SpaceX plans to begin launching a new generation of direct-to-device satellites next year. The company is targeting the first half of 2028 for commercial service.

Direct-to-device technology allows standard mobile phones to communicate with satellites without requiring a Starlink dish.

The current version largely supports text messages and emergency communications in places without cellular coverage. SpaceX intends the next generation to provide service comparable to 5G, including broader data capabilities and global roaming.

The company has acquired mid-band wireless spectrum from EchoStar in the United States. Spectrum consists of the licensed radio frequencies used to carry mobile signals.

Owning spectrum gives SpaceX several options. It could offer mobile service directly, build terrestrial infrastructure to supplement its satellites or form partnerships with existing wireless carriers.

International expansion will require regulatory approvals in individual countries. That process is likely to resemble Starlink’s broadband rollout, which expanded gradually as SpaceX negotiated with national regulators.

What investors should watch

The $1.11 billion-a-month hosting agreement makes AI infrastructure a substantial near-term business for SpaceX rather than a distant ambition.

Third-party customers can help finance the company’s data centers while SpaceX develops proprietary AI products and prepares to deploy computing equipment in orbit. If management’s stated payback periods are achieved, the hosting business could also reduce the financial risk of its aggressive expansion.

The strategy still depends on several demanding assumptions. SpaceX must obtain enough power and Nvidia processors, renew or replace short-term hosting contracts and demonstrate that its own AI products can monetize computing capacity effectively.

Its more ambitious goals require Starship to become reliable and fully reusable. Orbital computing must then prove that its savings in electricity, cooling and real estate outweigh the costs and operational difficulties of placing data centers in space.

The milestones for investors include:

* Revenue recognition from the new hosting agreement beginning Dec. 1* Progress toward the $100 billion ARR run rate* Expansion from roughly two gigawatts to five to 10 gigawatts of terrestrial computing capacity* Continued access to Nvidia processors* Renewal or replacement of six-month hosting commitments* Regular Starship flights carrying commercial payloads* Recovery and reuse of both Starship stages* Initial orbital-computing launches* Deployment of direct-to-device satellites ahead of the planned 2028 service

Johnsen said orbital computing remains the least understood part of the company because outside observers expect it to arrive much later than SpaceX does.

“The piece that’s still a little misunderstood is orbital compute,” he said.

For now, however, the new hosting contract gives investors a more immediate number to consider: $1.11 billion a month from a single customer in a market where demand for computing capacity continues to exceed supply.