By Lina Irawan 
The scale of SpaceX’s new venture is breathtaking. In May, the company solidified a landmark agreement to provide Anthropic, a leading AI research company, access to an estimated 325,000 Nvidia GPUs housed within its Colossus data centers. This monumental deal is set to generate an eye-watering $1.25 billion per month for SpaceX. Weeks later, the company secured another massive contract, this time with tech giant Google, committing approximately 110,000 GPUs for a staggering $920 million per month. Combined, these two contracts alone project an annual revenue stream of roughly $26 billion – a figure that not only dwarfs SpaceX’s entire revenue from the previous year but also fundamentally reshapes its financial outlook.
This unexpected foray into high-performance computing infrastructure-as-a-service (IaaS) provides a tangible explanation for SpaceX’s stratospheric valuation. While Musk’s ambitious visions for building orbital data centers and establishing a self-sustaining Mars colony remain captivating long-term goals, they are also years, if not decades, away from generating substantial revenue. These colossal compute deals, however, offer immediate, substantial cash flow from infrastructure the company has already invested in and built, bridging the gap between futuristic aspirations and present-day financial realities.
According to Sean Cray, a senior analyst covering telecom, media, and technology at Moody’s, this new business model is a masterstroke. “It gives SpaceX immediate revenue from infrastructure it has already built, while giving investors another reason to view the company as more than a rocket manufacturer, as Musk has repeatedly positioned it,” Cray noted. This diversification not only de-risks SpaceX’s portfolio from the inherent volatilities of space launches and satellite operations but also positions it at the nexus of the most transformative technological shift of our era: artificial intelligence.
The business model thrives on a fundamental imbalance in the current tech landscape: demand for AI compute capacity continues to vastly outpace supply. As the global AI race intensifies, companies are scrambling to develop and deploy increasingly powerful large language models (LLMs) and other advanced AI applications. This requires an astronomical amount of processing power, primarily from specialized graphics processing units (GPUs) like those manufactured by Nvidia. Building and scaling a large-scale data center capable of housing tens or hundreds of thousands of these advanced GPUs, complete with robust power infrastructure, sophisticated cooling systems, and high-speed networking, can demand years of development, billions in capital investment, and specialized expertise.
This acute shortage makes SpaceX’s offering incredibly attractive. By leveraging its existing, massive Colossus infrastructure, SpaceX can provide immediate, high-end computing power to customers who desperately need it now. This scarcity allows SpaceX to charge a significant premium, as Cray highlighted to Fortune, capitalizing on the urgency of AI developers to accelerate their model training and deployment. The effective cost per GPU per month in these deals is indicative of this premium. For Anthropic, 325,000 GPUs for $1.25 billion a month translates to approximately $3,846 per GPU per month. For Google, 110,000 GPUs for $920 million a month equates to roughly $8,363 per GPU per month. While these are broad averages and actual pricing can vary based on specific GPU models, usage patterns, and contractual terms, they underscore the immense value placed on immediate access to such critical resources.
Last year, SpaceX’s revenue reached $18.7 billion. The vast majority of this came from its Starlink satellite internet service, which contributed $11.4 billion, and its traditional launch business, including Falcon 9 and Falcon Heavy missions, and early Starship development, which accounted for about $4.1 billion. The company’s nascent AI segment, primarily driven by its xAI venture and the development of the Grok large language model, contributed a modest $3.2 billion. However, this segment also recorded a substantial operating loss of approximately $6.4 billion, even as Elon Musk has consistently touted AI as SpaceX’s next major money-maker, particularly after the acquisition of xAI in February.
The significant operating loss in the AI segment is not entirely unexpected. Developing and training a sophisticated large language model like Grok demands massive upfront investment in GPUs, the immense electricity required to power them, and the complex infrastructure to support their operation. Furthermore, Grok, while promising, has so far lagged behind established models created by competitors like Anthropic and OpenAI in terms of both revenue generation and advanced capabilities. The compute renting business, however, offers a pragmatic and proven path to profitability for SpaceX’s AI segment, irrespective of Grok’s immediate commercial success.
"It shows that there’s just different pathways for them to generate revenue in their AI segment," Cray elaborated. "It doesn’t strictly have to come from Grok and their AI enterprise applications." This flexibility is a key advantage. SpaceX maintains complete discretion over its hardware resources. It can rent out capacity to external customers like Anthropic and Google, but it can also dynamically reallocate that compute power to train newer, more capable versions of Grok, or even redirect it to support Starlink’s own internal computational needs, such as optimizing satellite constellation management or processing vast amounts of telemetry data.
However, this inherent flexibility introduces a potential pitfall, as warned by Sridhar Tayur, a professor of operations management at Carnegie Mellon University. Built into the contracts with Google and Anthropic is a crucial 90-day cancellation provision. This clause grants SpaceX the unilateral ability to reclaim its compute resources with relatively short notice if Grok’s training needs or other internal demands suddenly surge. While this provides SpaceX with strategic agility, the same clause also empowers its customers to walk away from the agreements if more affordable or more suitable compute capacity becomes available elsewhere. This raises questions about the long-term stability and predictability of the revenue generated from this business.
"Is it a one-off thing that is not your main line of business?" Tayur pondered, "Or is infrastructure as a service going to become a permanent part?" The answer to this question will largely depend on the persistence of the AI compute shortage and the evolving competitive landscape.
Indeed, SpaceX’s competitors have already taken note of the lucrative potential that comes with monetizing excess compute capacity. Meta, the parent company of Facebook and Instagram, is reportedly in advanced discussions to lease its computing power to Anthropic in a deal that could be worth up to $10 billion over two years, as reported by Reuters and the New York Times. This demonstrates that other tech giants, having made significant investments in AI infrastructure for their own needs, are now exploring similar avenues to monetize their assets.
Despite these potential contractual volatilities and emerging competition, SpaceX’s prospects in this domain appear bright for now. The Wall Street Journal recently reported that SpaceX is in active discussions to provide the Pentagon with data-center capacity, potentially worth billions of dollars, for running advanced AI models. This potential deal, if finalized, would significantly strengthen the argument that selling compute power could evolve into a foundational, long-term business for SpaceX, rather than a mere temporary revenue stream. A contract with a demanding, long-term client like the U.S. government would lend immense credibility and stability to this new venture.
The very existence of the Colossus data center complex, and thus SpaceX’s ability to sell its compute, is a testament to Elon Musk’s characteristic foresight and aggressive execution. Back in 2024, xAI, Musk’s AI venture, brought its first major Colossus cluster online in an astonishing 122 days by rapidly converting an existing factory. This remarkable feat showcased SpaceX’s unique ability to deploy large-scale infrastructure at an unprecedented pace. Since then, the company has aggressively expanded its Memphis-area complex to roughly 2 million square feet across its Colossus and Colossus II facilities. These complexes are designed to provide approximately 1 gigawatt of compute power, with ambitious plans to eventually house 1 million GPUs. This rapid, massive build-out was initially driven by the need to train Grok, but the sheer scale inadvertently created an invaluable asset for external monetization.
SpaceX’s colossal $1.8 trillion valuation hinges on the expectation of enormous future growth and the successful realization of audacious, capital-intensive projects. Critics have long argued that this market value is heavily dependent on long-shot bets—like fully reusable Starship missions, large-scale Mars colonization, and a globally ubiquitous Starlink network—that would require massive, sustained investments over many years before generating meaningful, consistent revenue, assuming they even pan out as planned.
The agreements with Anthropic and Google, however, fundamentally alter this narrative. They demonstrate that SpaceX can generate billions of dollars in immediate, high-margin revenue from sophisticated infrastructure it already owns, providing a critical financial runway while it continues to work toward those inherently risky, long-term goals. This financial injection reduces reliance on constant capital raises or debt, offering greater stability for its core space ventures.
The lasting viability of this compute renting business will ultimately depend on whether the global shortage of AI infrastructure persists. Major players are working to address this bottleneck. OpenAI, for instance, in collaboration with Microsoft, announced the ambitious "Stargate" initiative last year, a project aiming to build a vast network of data centers across the U.S. with a potential price tag of up to $100 billion, designed to house millions of GPUs and provide multi-gigawatt power capacity to aid future AI development. Such initiatives, if successful, could eventually ease the supply constraints currently driving up compute prices.
Yet, for the foreseeable future, SpaceX holds a unique and powerful position. It possesses something nearly every major AI developer desperately wants and cannot easily build fast enough or affordably enough on their own. The combination of cutting-edge Nvidia GPUs, a massive, rapidly deployed data center footprint, and a flexible commercial approach positions SpaceX as a critical enabler in the ongoing AI revolution. As Sean Cray aptly concluded, "There will probably be other third parties that would be interested in that compute because pretty much at this rate it’s a race to secure that compute capacity." SpaceX has not just entered this race; it has become a major supplier of the very fuel that powers it.
