Anthropic will pay xAI $1.25 billion per month for compute
Original reporting by TechCrunch
Earlier this month, Anthropic sent ripples through the AI world with a surprising move: a deal to acquire 300 megawatts of compute. This massive purchase, securing the entire output of the Colossus 1 data center near Memphis, Tennessee, comes with an equally staggering price tag. Details emerging from SpaceX’s S-1 filing with the SEC reveal Anthropic will pay xAI an astonishing $1.25 billion per month through May 2029, a transaction that could funnel over $40 billion in revenue to xAI.
SpaceX, xAI’s parent, presented the arrangement as a strategic maneuver to “monetize unused compute capacity.” This emerging "neocloud" model allows AI companies to offset infrastructure costs by temporarily acting as cloud providers, utilizing spare capacity when their own consumption is low. For xAI, this involves selling excess server power, with the company stating it anticipates "additional similar services contracts."
The actual story
However, the subtext of this monumental agreement tells a more nuanced tale. Despite the rhetoric of a savvy dual monetization strategy, xAI appears to have significantly overbuilt its compute infrastructure. The dramatic decline in usage for Grok, xAI’s flagship AI assistant, has reportedly freed up substantial server resources. In a strategic pivot, xAI is now offloading this surplus capacity to a direct competitor, Anthropic, suggesting a scramble to monetize assets ahead of a potential public offering.
The monumental compute deal between Anthropic and xAI, potentially valued at $40 billion, transcends a simple transaction; it lays bare the immense capital expenditures involved in the AI race and the complex dance between ambitious infrastructure build-outs and real-world demand. While xAI frames its "dual monetization strategy" as savvy resource management, the underlying reality—a surplus of compute capacity resulting from declining Grok usage—underscores the inherent volatility in projecting future AI needs. This arrangement effectively transforms a direct competitor into a crucial revenue stream, a pragmatic if unusual move that highlights the evolving nature of competition and collaboration in a fiercely contested market. It's a vivid illustration of how even the most well-funded players must adapt to capitalize on their massive investments.
Shifting Compute Dynamics
Beyond the immediate financial implications, this deal signals a potential inflection point for the broader AI industry. The "neocloud" model, where leading AI developers become ad hoc compute providers, could become a common strategy for managing the prohibitive costs and risks associated with building vast data centers. This dynamic suggests a future where compute resources are increasingly fluid, bought, sold, and shared between even rival companies to optimize utilization and mitigate overbuilding. It blurs the lines between AI developer, infrastructure provider, and competitor, potentially fostering a more interconnected, albeit complicated, ecosystem. For the wider market, this trend underscores the pressing need for flexibility and strategic foresight, influencing how all players, from startups to established giants, approach their foundational compute investments in a domain where technological leaps and user adoption rates are anything but predictable. This fluidity marks a notable step towards a more mature, and perhaps rationalized, compute economy within AI.