When SpaceX and xAI completed their merger in February 2026, valuing the combined entity at $1.25 trillion, the tech world speculated about what the union would ultimately mean. Would xAI’s Grok models improve with SpaceX’s engineering talent? Would Starlink become the backbone of some grand AI deployment network? The answer, it turns out, was far more immediate and lucrative than most expected: SpaceX-xAI has quietly transformed itself into the most powerful AI compute landlord on the planet — and the leases it has signed in the months since are rewriting the economics of the entire industry.
A Data Center Empire Built on 330,000 GPUs
At the center of this story is Colossus 1, SpaceX’s sprawling data center complex in Memphis, Tennessee. Housing more than 220,000 Nvidia GPUs and running on 300 megawatts of power capacity, Colossus 1 represents one of the largest single concentrations of AI compute on Earth. When xAI built it, the intent was to fuel Grok’s development. But following the merger, SpaceX’s leadership made a striking strategic pivot: instead of consuming all that power internally, they would rent it out to the highest bidders — and the bidders came quickly.
Anthropic, the AI safety company behind the Claude family of models, struck the first deal. Under an agreement that runs through May 2029, Anthropic committed to paying $1.25 billion per month for exclusive access to the entirety of Colossus 1’s compute. At that rate, the total commitment stretches to roughly $50 billion over the life of the contract. For a company that had been publicly valued at under $1 trillion just months earlier, agreeing to spend that kind of money on raw compute is a remarkable signal of confidence in its own revenue trajectory — and a pointed bet that AI model training and inference demand will remain voracious for years to come.
But SpaceX wasn’t finished. Rather than letting Anthropic’s exclusive deal define the ceiling of its infrastructure ambitions, the company began assembling a second, separate pool of GPU capacity outside Colossus 1: approximately 110,000 Nvidia GPUs, with CPUs, memory, and associated hardware. The target customer this time was not a scrappy AI startup but the largest company in the world by market cap: Google.
Google’s $30 Billion Bridge
In a filing disclosed on June 5, 2026, Google confirmed it has agreed to pay SpaceX $920 million per month from October 2026 through June 2029 — a total commitment of approximately $30 billion — for access to that second GPU cluster. The deal came with an urgency clause that speaks to just how strained AI infrastructure has become across the industry.
A Google Cloud spokesperson explained the rationale with disarming candor: the deal was made “to ensure we have bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected.” The word “bridge” is doing heavy lifting there. Google is not renting SpaceX’s compute because it lacks the engineering vision to build its own; it is renting it because building takes time, and the demand for AI inferencing is not waiting. Customers want Gemini agents today, not in 2028 when the next Google data center campus comes online.
The contract terms reveal Google’s leverage as well as SpaceX’s. If SpaceX cannot deliver the committed GPU capacity by September 30, 2026, Google may terminate the agreement after a one-month grace period, or accept the available hardware at proportionally reduced cost. Both parties retain the right to walk away with 90 days’ notice after December 31, 2026. These are not the terms of a company locking in a desperate buyer; they are the terms of a market where both sides understand exactly what GPU hours are worth.
The $75 Billion Runway
Add the two contracts together and the arithmetic is staggering. In less than four months since the xAI merger closed, SpaceX has signed compute-lease agreements representing a combined $75 billion in future contracted revenue. To put that in context: it took Amazon more than a decade to build AWS to comparable annual revenue figures. SpaceX, by repurposing a data center that was originally built to train a chatbot, has essentially created a cloud computing business in one fiscal quarter.
That number carries particular significance given what is coming next. SpaceX has filed its S-1 and is targeting an IPO on June 12, 2026, seeking to raise $75 billion at a company valuation of $1.75 trillion. Bankers rarely design coincidences of this magnitude by accident. The back-to-back announcement of the Anthropic and Google deals — each providing a near-identical dollar figure in contracted future revenue — gives institutional investors something precious: visibility. In a market skeptical of AI’s ability to translate hype into durable cash flows, $75 billion in signed contracts from two of the most creditworthy customers imaginable is as close to certainty as venture-stage infrastructure gets.
Why Everyone Is Renting From Musk
The deeper question the Anthropic and Google deals raise is a structural one: why are two of SpaceX’s fiercest ideological competitors writing it nine-figure monthly checks?
The answer is the GPU shortage, which despite years of Nvidia production ramp-ups, has not meaningfully abated. Training frontier AI models requires tens of thousands of high-end GPUs operating in unison, and building the networking fabric and power infrastructure to run them at scale takes years. SpaceX had a significant head start through xAI’s early investment in Colossus 1, and the result is that it holds one of the few large, immediately available pools of GPU capacity that isn’t already spoken for by a hyperscaler.
There is also something philosophically uncomfortable, though publicly unacknowledged, about the situation Anthropic finds itself in. The company was founded in 2021 on the explicit premise that developing AI safely required organizational independence from the ambitions of any single powerful actor. It has received billions in investment from Google and Amazon partly to protect against dependence on any one patron. And yet, month by month, it will now be writing $1.25 billion checks to a company controlled by Elon Musk — a figure whose AI philosophy, expressed through Grok and through xAI’s rapid scaling, diverges sharply from Anthropic’s own. The compute economy, it appears, does not care much about philosophical alignment.
A New Kind of Infrastructure Giant
What SpaceX is building through these deals is something genuinely new. It is not a traditional cloud provider — it lacks the global distribution, the software services stack, and the enterprise sales motion that define AWS, Azure, and Google Cloud. It is not purely an AI lab, since the models xAI trains will increasingly compete with the same customers renting its hardware. It is closer to a GPU utility: a provider of raw compute at planetary scale, with the financial heft and technical credibility to attract commitments from companies that could theoretically build their own capacity.
Whether that model holds once SpaceX’s own IPO proceeds start flowing into new data center construction remains to be seen. A company valued at $1.75 trillion will face intense pressure to grow, and the fastest path to growth in AI infrastructure is to own more of the stack — not just the GPUs, but the networking, the software, the models. The day may come when SpaceX-xAI is no longer content to be Google’s and Anthropic’s landlord. On that day, the terms of today’s compute leases will look very different.
For now, though, the numbers speak for themselves. In a world where everyone needs AI compute and very few have enough, SpaceX has turned its early bet on xAI into one of the most lucrative landlord positions in the history of technology. The rent, for the moment, is extraordinary.
Sources: CNBC, TechCrunch, Bloomberg, Data Center Dynamics
