Overlooked detail in SpaceX's prospectus: Anthropic spends $15 billion a year renting Musk's GPUs

```

While everyone is intently watching SpaceX aiming for the largest IPO in history with a $1.75 trillion valuation, a detail in the prospectus that has been overlooked reveals a deeper signal: the bottleneck in the AI race is no longer the models, but computing power.

SpaceX’s prospectus shows that Anthropic has committed to pay SpaceX $1.25 billion per month, amounting to about $15 billion annually, for GPU computing power from its Colossus and Colossus II data centers. These two data centers have a combined computing power exceeding 1 gigawatt and were originally built mainly to serve xAI and its Grok chatbot.

This partnership also means that SpaceX is performing a delicate capital balancing act: the cash flow contributed by Starlink is being swallowed by xAI’s black hole of computing power consumption, and the idle GPU resources are then leased to other AI competitors. SpaceX has become an AI “shovel seller.”

SpaceX’s Computing Power Ledger: Self-use with the Left Hand, Leasing with the Right

The prospectus indicates that Anthropic will pay an undisclosed discounted fee in May and June, after which full monthly payments of $1.25 billion will begin, continuing until May 2029. As xAI's largest external customer, Anthropic’s contract is worth about $40 billion in total, making up the core revenue support for xAI’s AI segment in 2026.

It’s worth noting that an investigation by The Information revealed a key risk: either party can cancel the agreement with 90 days’ advance notice. This highly unusual clause means that the $40 billion contract could disappear within 90 days.

The Colossus and Colossus II data centers span Tennessee and Mississippi, equipped with around 100,000 H100 chips and 220,000 GB200/GB300 chips, making them one of the largest single AI computing clusters in the world. SpaceX originally fast-tracked the facilities for its xAI division. Elon Musk later stated that SpaceX ultimately does not need all of the computing capacity and revealed that it is in discussions with other companies for similar arrangements.

SpaceX said in its S-1 filing that it expects to sign more computing service contracts while continuing to use the data centers for its own operations. The company stated that its computing capacity is sufficient to meet the demands of both its own AI model training and reasoning, as well as external agreements.

In Q1 2026, out of SpaceX’s $10.7 billion capital expenditure, $7.7 billion went to AI—76% of the total. This means that most of Starlink’s earnings have been invested into computing power. But now, this $40 billion order shows that SpaceX has redefined this model as a “dual monetization strategy,” claiming it provides multiple paths for capital returns. In other words, SpaceX is turning its AI infrastructure from an internal cost center into an asset capable of generating external revenue even before the IPO.

Anthropic’s Computing Power Spending Proposition

What does an annual computing power expenditure of $15 billion mean for Anthropic?

Anthropic’s revenue in Q1 was $4.8 billion, and Q2 is expected to break $10 billion, with a growth rate even outpacing Google and Facebook on the eve of their IPOs. Even so, annual computing power spending of $15 billion takes a significant share of its yearly income. The market’s concern is this: Is Anthropic pouring its rapidly growing revenue continuously into computing power, without forming sustainable profit accumulation?

Revenue and computing costs skyrocketing in tandem constitute the most delicate growth paradox facing AI companies today. On one hand, applications like AI programming tools are booming, triggering rapid expansion in model training and inference—and a surge in demand for computing power. On the other, while Anthropic’s single-quarter leap from $4.8 to $10.9 billion is eye-catching, it’s hard to ignore one fact: the faster the growth, the greater the pressure to procure computing resources. Whether this “growth for growth’s sake” cycle can persist depends on whether the company can find a true balance between investment in computing and business returns.

As computing power itself becomes a more scarce strategic resource than capital, the core competitive dimension of the AI industry may quietly shift in the future—not just a race of model capabilities, but a race to lock in sufficient GPUs, electricity, and data centers ahead of the curve, enabling smoother progress on this high-consumption track.

Risk Warning and DisclaimerThe market has risks and investment requires caution. This article does not constitute individual investment advice and does not take into account any user’s specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their own circumstances. Investing accordingly is at your own risk. ```