A $13 billion run-rate deal, not a conventional space contract
SpaceX has moved another step into the AI infrastructure race with a newly disclosed hosting contract expected to generate about $1.11 billion per month starting December 1, 2026 . On an annualized basis, that works out to roughly $13.3 billion, a figure that explains why the announcement immediately became a focal point for investors tracking SpaceX’s transformation from a launch-and-connectivity company into a broader space, cloud and artificial intelligence platform .
The disclosure came from SpaceX chief financial officer Bret Johnsen during the Goldman Sachs Communacopia + Technology Conference on September 10, 2026 . Johnsen said the agreement was closed earlier in September and that it gave management more confidence in SpaceX’s goal of reaching a $100 billion annual recurring revenue run rate by the end of 2026 . GuruFocus separately reported the same monthly value and said the contract is scheduled to start on December 1 .
The customer has not been named . That is important because the identity of the counterparty would help investors judge the quality, duration and strategic value of the agreement. For now, the market has a large number, a start date and management’s explanation that the deal fits into SpaceX’s rapidly expanding AI hosting business .
The contract is also not a traditional aerospace order. It is not a NASA launch award, a Starlink government services package or a satellite deployment contract. It is an AI compute hosting agreement, meaning SpaceX is renting out infrastructure capacity — power, facilities, processors, cooling and related systems — to a customer that needs large-scale compute . That is why the story matters: a company best known for rockets is now describing compute capacity as a major driver of its financial trajectory .
Why the timing matters
The December 1 start date is central to the financial framing . Johnsen described SpaceX’s $100 billion ARR target as a December exit-rate calculation, not as revenue already earned across 2026 . In other words, management is annualizing the level of business expected at year-end rather than saying SpaceX has already booked $100 billion in trailing twelve-month sales.
That distinction is more than accounting vocabulary. It determines how investors should interpret the $13.3 billion headline. If the new AI hosting contract starts as planned and bills at the stated monthly rate, it becomes a major contributor to SpaceX’s December run rate . But it does not automatically mean SpaceX will collect $13.3 billion from that customer during calendar 2026, because the agreement begins in the final month of the year .
Investing.com reported that William Blair reiterated an Outperform rating after the announcement and described the new SpaceXAI compute deal as worth $1.11 billion per month, or $13.3 billion annualized . The same report said SpaceX confirmed that the new deal is separate from a previously announced $6.7 billion six-month agreement discussed on its August earnings call . That separation matters because it suggests the latest contract adds incremental demand rather than simply restating an earlier customer commitment .
The immediate strategic message is clear: SpaceX is trying to prove that AI compute can become a scaled commercial business alongside launch services and Starlink connectivity . The financial question is whether the company can convert annualized demand into durable recurring revenue while continuing to fund the enormous infrastructure buildout required to serve AI customers.
The six-month structure is the key caveat
The biggest caution around the announcement is the contract’s duration. In the conference transcript, Johnsen said SpaceX’s compute deals have generally been structured as roughly “90-day with a 90-day out,” or about six-month commitments . He said the newly discussed deal follows the same structure and timeline .
StockTi highlighted this point as the main risk behind the headline, noting that a six-month minimum structure limits long-term visibility even though the annualized revenue number is striking . That is the difference between a $13.3 billion run-rate agreement and a long-term take-or-pay contract. One signals demand and pricing power; the other would provide stronger certainty about future cash flows.
SpaceX appears to have chosen shorter contracts deliberately . Johnsen said the company wants to avoid locking away too much compute capacity if its internal AI products need that capacity later . That logic fits the broader SpaceXAI strategy: monetize infrastructure externally today, but preserve flexibility to redirect capacity to Grok, Cursor or other in-house AI products if their economics become more attractive .
This flexibility cuts both ways. It can protect SpaceX from underpricing compute in a fast-moving market, but it also means the customer can potentially leave after a short period . For investors, the question is not only whether SpaceX can sign enormous AI hosting deals. It is whether those customers renew, expand and remain profitable over time.
AI infrastructure becomes part of SpaceX’s vertical-integration story
Johnsen placed the new contract inside SpaceX’s familiar vertical-integration narrative . He argued that SpaceX historically built its own rocket components because outside suppliers could not move at the company’s required pace or quality level, then applied a similar model to Starlink by owning launch, satellites and customer relationships . The AI business, he said, follows the same pattern: SpaceX is building infrastructure, arranging power and using its own models to reach customers .
That framing is designed to make AI compute feel like a natural extension of SpaceX rather than a side business. In launch, SpaceX reduced costs and increased cadence through reusable rockets. In Starlink, it combined manufacturing, launch and network operations. In AI hosting, management is now arguing that similar control over facilities, power and deployment speed can create an advantage in a market constrained by supply .
StockTwits reported that Johnsen described SpaceX’s relationship with Nvidia as “very strong” and said the company is relying on Nvidia hardware as it expands terrestrial AI compute capacity . The report said SpaceX plans to end 2026 with a little over 2 gigawatts of terrestrial compute and aims to deploy between 5 and 10 gigawatts in 2027 . Those figures, if achieved, would represent a massive infrastructure ramp and explain why power availability, permits and processor allocation are now central to SpaceX’s AI story .
The Nvidia relationship is another key dependency. Johnsen said SpaceX had talked publicly about being Nvidia-exclusive and hoped that relationship would help with compute allocation, while acknowledging that processor supply is one piece the company does not fully control . That means execution depends not only on SpaceX’s construction speed but also on access to advanced chips.
The space angle: terrestrial compute now, orbital compute next
The milestone is not only financial. It also connects directly to SpaceX’s longer-term plan to merge space systems and AI infrastructure. Johnsen said SpaceX expects to launch its first computing satellites next year and sees orbital compute as compelling because power, cooling and permitting are becoming constraints on Earth . In the conference transcript, he described orbital compute as a potential solution to the industry’s pressure points around power, cooling and regulatory approvals .
That is the bridge between the “space” and “AI” parts of the story. The current contract appears to be tied to terrestrial hosting capacity, but management is already presenting orbital infrastructure as the next stage . If SpaceX can use Starship to deploy compute satellites at scale, the company could attempt to extend its Starlink playbook from communications bandwidth to processing capacity.
The ambition is still highly execution-dependent. Orbital data centers would need to solve issues around launch economics, reliability, heat management, data transfer, maintenance and customer trust. But SpaceX’s ability to use its own launch system gives it a distinctive route to test the concept . Johnsen also linked Starship’s development to the broader business, saying the vehicle is core to SpaceX’s future across multiple segments .
For now, the new $13.3 billion annualized contract is a terrestrial AI hosting win. Strategically, however, SpaceX is using it to validate a broader thesis: demand for compute is so intense that the company’s space infrastructure, power strategy and AI software ambitions may belong in one integrated business model .
What investors should watch next
The first item to watch is whether the contract begins billing on December 1, 2026 at the stated $1.11 billion monthly rate . The second is whether SpaceX discloses the customer or provides enough detail for investors to assess counterparty quality . The third is renewal behavior: a six-month structure can still become valuable recurring revenue if customers keep extending capacity agreements .
Capacity execution is equally important. SpaceX says it has visibility into the power, facilities and permitting needed to support its 2027 deployment range . If the company misses those buildout targets, AI hosting revenue could fall short of the annualized figures now being discussed. If it hits them, the new contract may look less like an isolated headline and more like evidence that SpaceX has found a second infrastructure flywheel next to Starlink.
The bottom line is that SpaceX has secured a major AI compute contract with a headline annualized value of about $13.3 billion . It strengthens the company’s claim that AI hosting can become a core revenue engine, but it also comes with caveats: an undisclosed customer, a short commitment period, heavy reliance on infrastructure execution and dependence on advanced chip supply . That combination makes the deal a milestone — and a test — for SpaceX’s attempt to fuse space operations with the economics of the AI boom.