Abstract:
SpaceX is rapidly expanding its artificial intelligence infrastructure business and has secured a staggering order for computing power. SpaceX Chief Financial Officer Brett Johnson recently revealed that the company signed a cloud computing agreement with an undisclosed customer earlier this month. Starting in December 2026, the customer will pay SpaceX approximately $1.11 billion per month for the right to use AI computing infrastructure. Based on this amount, the annualized value of the contract will reach approximately US$13.3 billion.
Johnson disclosed the above information at a technology industry conference held by Goldman Sachs, but he did not disclose the name of the client or the specific duration of the contract. However, judging from the amount alone, this is enough to become one of SpaceX’s largest AI computing power contracts to date. With this new contract starting to count in December, SpaceX is on track to achieve its $100 billion annual recurring revenue run rate goal by the end of 2026, Johnson said.
This is not the first time SpaceX has sold large-scale AI computing power to external companies. Since this year, the company has successively signed huge computing power agreements with a number of large AI companies. Among them, Anthropic previously signed a monthly contract of approximately US$1.25 billion; Google signed a monthly agreement of approximately US$920 million, starting in October 2026, to obtain approximately 110,000 Nvidia GPUs, as well as supporting CPUs, memory and other computing resources.
In addition, AI startup Reflection AI has also become a SpaceX customer. The company previously signed a monthly contract of approximately US$150 million to obtain Nvidia GB300 computing resources deployed at the SpaceX Colossus 2 data center near Memphis, Tennessee, USA.
If this mysterious customer’s contract is calculated together with the agreements with Anthropic, Google and Reflection AI, SpaceX currently has a monthly contract amount of approximately US$3.4 billion from these disclosed or confirmed AI computing power contracts alone, corresponding to an annualized scale of more than US$41 billion. This means that AI infrastructure is rapidly growing from outside SpaceX’s traditional business into a large-scale independent revenue source.
SpaceX’s rapid entry into the AI infrastructure market is closely related to the large-scale GPU data centers the company has built in recent years. Its core facilities include the Colossus and Colossus 2 data centers in Memphis. SpaceX deploys Nvidia GPUs on a large scale in these facilities and then sells some of its computing resources to other companies through cloud computing agreements.
This business model allows SpaceX to play the role of AI infrastructure builder and cloud computing provider at the same time. On the one hand, companies need to build huge computing clusters for their own AI projects; on the other hand, they can rent out computing power that is not currently used by their own projects to other AI companies, thereby increasing the utilization of expensive GPU infrastructure.
SpaceX has generated $1.6 billion in new AI infrastructure revenue from cloud service agreements in the second quarter of 2026. The company's AI business revenue reached US$2.6 billion in the quarter, a year-on-year increase of 247% and a month-on-month increase of 213%. At the same time, the adjusted EBITDA of the AI business became profitable for the first time, reaching US$1.1 billion.
As of the end of the second quarter of 2026, SpaceX already has approximately 1.4GW of computing power, a significant increase from approximately 1GW in the first quarter and 400MW a year ago. The company expects computing capacity to exceed 2GW by the end of 2026.
SpaceX CEO Elon Musk also previously stated that the company plans to further accelerate the construction of computing infrastructure. According to current plans, by the end of 2027, SpaceX’s cumulative online computing capacity may be close to 10GW, much higher than the previous target of approximately 5GW.
Such a radical expansion requires huge capital investment. In the second quarter of 2026, SpaceX's capital expenditures were approximately US$18.4 billion, of which approximately US$15.8 billion was directly used for AI computing infrastructure, and the remaining funds were mainly used for starship R&D and production, launch facilities, satellite production, and global ground station expansion.
The reason why SpaceX is willing to invest such a huge amount of money is closely related to the current shortage of AI computing power market. With the rapid development of large-scale language models and AI agents, the number of GPUs required by AI companies continues to increase, but data center construction, power supply, and high-end GPU delivery speeds are difficult to meet the demand simultaneously.
SpaceX management believes that this supply and demand imbalance will not end in the short term. Johansen previously stated that the newly signed cloud computing agreement is bringing very favorable economic benefits, and the company's newly deployed AI computing infrastructure can currently even recover the investment cost in less than a year.
This is also an important reason why SpaceX continues to expand investment in AI infrastructure. Traditional data centers usually take many years to recoup the huge construction costs, but if AI computing power can be rented for a long time at the current market price, then building a GPU data center may become a capital-intensive business with a very short return cycle.
However, SpaceX has not reduced its demand for AI computing power. The company still requires significant computing resources to train and run Grok. Musk said that future Grok models will further utilize the large amount of engineering and operational data accumulated by SpaceX over the past few decades, including data from rockets, satellites and other aerospace systems, thereby strengthening the model's capabilities in the engineering field.
SpaceX also plans to further use NVIDIA's latest generation computing platform to build AI infrastructure. Musk said that the company will expand future AI computing capabilities based on Nvidia's Vera Rubin architecture and believes that this platform is very suitable for SpaceX's future AI needs. Therefore, the new computing power can be used for Grok's own training, or it can be sold to external customers as commercial cloud computing resources.
More ambitiously, SpaceX is preparing to send AI computing power into space. The company has proposed plans to build an orbital data center and expects to launch related computing satellites as early as 2027. Musk has previously emphasized that the orbital data center is not a distant long-term vision, but an actual project that SpaceX is advancing.
According to current plans, these orbital computing platforms will use the NVIDIA Vera Rubin NVL72 architecture optimized for the space environment. SpaceX believes that this highly integrated computing platform is more suitable for deployment in an orbital environment than traditional rack-mounted data centers.
SpaceX believes that deploying AI computing infrastructure into space has certain potential advantages. Traditional large-scale data centers require large amounts of land, power and cooling resources, while orbiting data centers can be powered by solar energy and use the space environment for heat dissipation. As AI computing needs continue to grow, this approach could become an important part of SpaceX's long-term AI infrastructure strategy.
However, before the orbital data center can truly become commercially viable, SpaceX still needs to solve a series of issues such as launch costs, equipment reliability, heat dissipation, communications, and the long-term operation of computing hardware in the space environment. Therefore, SpaceX’s current commercial revenue from AI computing power still mainly comes from ground data centers.
Judging from the current development speed, AI computing is becoming another important growth point for SpaceX after launch services and Starlink. On the one hand, the company has the huge cash flow and infrastructure capabilities brought by Starlink, and on the other hand, it has the world's leading rocket launch capabilities, which enables it to form an advantage that other AI cloud service providers are difficult to copy in terms of building AI data centers and deploying orbital computing facilities in the future.
This monthly mystery customer contract of US$1.11 billion further proves that the market demand for large-scale AI computing power is still extremely strong. SpaceX has not disclosed the identity of the customer or the length of the contract, so it is currently unclear how much revenue this order will ultimately contribute to the company.
But looking at the monthly amount alone, the contract generates more revenue each month than many technology companies make in an entire year. As large AI companies such as Anthropic and Google continue to compete for computing resources, and SpaceX continues to expand the scale of its Colossus series of data centers, AI computing infrastructure is quickly becoming a new growth engine for SpaceX that may rival its traditional core business.
If SpaceX can ultimately achieve an annual recurring revenue run rate of US$100 billion by the end of 2026 as planned, the AI computing power business will become an important part of promoting this goal. At the same time, as ground-based GPU data centers continue to expand and gradually extend to orbital computing platforms, SpaceX is trying to establish a complete AI computing system that can be controlled by itself from GPU computing power, data centers to space infrastructure.
Comments