IPO prospectus: Anthropic has agreements to pay SpaceX up to $84.5B to use its Nvidia-based compute through 2029, mostly cancellable with a 90-day notice period

The cost of foundational AI model training and inference infrastructure increases significantly for leading AI labs.

Anthropic has entered into agreements to pay SpaceX up to $84.5 billion for computing capacity, primarily utilizing Nvidia's hardware, through the year 2029. The deal grants Anthropic access to SpaceX's significant computational resources. A key feature of these agreements is a provision allowing for cancellation with a 90-day notice period, offering Anthropic flexibility in managing its long-term infrastructure commitments. This substantial financial commitment highlights Anthropic's aggressive expansion in AI development and its reliance on high-performance computing, with SpaceX emerging as a significant infrastructure provider in this domain.

This agreement signifies a major shift in how AI companies are securing critical compute resources, moving towards massive, long-term contracts with specialized providers rather than relying solely on hyperscalers. The sheer scale of the $84.5 billion figure underscores the immense capital expenditure required to train and operate state-of-the-art AI models. It also positions SpaceX, traditionally known for space exploration, as a formidable player in the AI infrastructure market, leveraging its significant investment in Nvidia hardware. This could signal increased competition and potentially more diversified options for compute-intensive businesses in the future.

The cancellable nature of the contract with a 90-day notice period provides Anthropic with crucial flexibility, mitigating the risk of being locked into expensive hardware commitments if its needs or technology landscape change rapidly. However, it also presents a potential challenge for SpaceX in ensuring consistent revenue and resource allocation. This move indicates a growing trend of AI companies seeking dedicated, high-density compute solutions tailored to their specific needs, potentially leading to a more specialized and tiered market for AI infrastructure services.