SpaceX has disclosed a landmark compute agreement with AI safety company Anthropic worth up to $40 billion over several years, surfaced through the aerospace company's initial public offering (IPO) filing on Wednesday. Under the arrangement, Anthropic is paying SpaceX $1.25 billion per month for access to data centre compute capacity, with payments running through May 2029 — though either party retains the right to exit the deal with just 90 days' notice.

What the IPO filing reveals

The deal's details emerged only because SpaceX is now pursuing a public listing, which requires the disclosure of material commercial contracts. According to the filing, the total potential value of the arrangement reaches $40 billion, making it one of the largest AI infrastructure agreements ever publicly documented. The $1.25 billion monthly payment figure underscores the extraordinary scale at which frontier AI labs are now procuring compute.

Why it matters

The agreement signals a significant diversification in how AI companies source raw computing power. Anthropic, founded in 2021 by former OpenAI executives Dario and Daniela Amodei, had previously announced multi-year cloud agreements with Amazon Web Services and Google Cloud for GPU capacity. Entering a separate, large-scale arrangement with SpaceX — primarily known for its launch vehicles and Starlink satellite constellation — points to how intensely AI labs are competing for every available source of infrastructure.

The competitive backdrop

The AI industry's insatiable appetite for compute has driven frontier labs into increasingly unconventional infrastructure partnerships. Traditional hyperscalers alone can no longer satisfy demand from the largest model developers, pushing companies like Anthropic to diversify across multiple providers simultaneously. The fact that a 90-day exit clause is built into this deal reflects the volatility of compute demand forecasting — training runs, inference scaling, and research priorities can shift rapidly.

SpaceX's expanding infrastructure play

SpaceX, founded in 2002 by Elon Musk, has steadily expanded beyond aerospace into data infrastructure, with its Starlink network representing a significant compute and connectivity asset base. Monetising data centre capacity through long-term agreements with AI companies represents a new and highly lucrative revenue stream as the company moves toward a public market debut. The Anthropic deal alone, at its maximum value, would represent a transformative commercial contract for any company at IPO stage.

What's next

Investors and analysts will now scrutinise the durability of the $1.25 billion monthly revenue stream, given the 90-day termination clause that either party can invoke. For Anthropic, the key question is whether its compute diversification strategy translates into a competitive advantage in model training and inference at scale. The broader AI infrastructure market will be watching closely to see whether other frontier labs follow suit with similarly large, non-hyperscaler compute commitments.