Anthropic has agreed to spend up to $120.5 billion leasing computing power from Elon Musk's SpaceXAI to run its artificial intelligence models.
The artificial intelligence developer disclosed the massive multi-year deal in documents filed for its initial public offering. The agreements run through 2029 and secure access to specialized computing clusters powered by Nvidia graphics processing accelerators.
Between May and October 2026, the two companies signed and amended several contracts for server capacity to support AI development. Anthropic expects SpaceXAI to deliver the full volume of contracted computing power by December 2026, linking the firm to the Colossus 2 supercomputer.
Once the infrastructure reaches full operational capacity, Anthropic expects payments to reach about $3.1 billion each month, or $37.2 billion per year. Until the deployment is fully operational, the price of the services will adjust as SpaceXAI brings additional computing hardware online.
Contract deadlines and cancellation rules
The initial mandatory commitment periods under the contracts expire in January and March 2027. Following the end of those early binding periods, either party will have the right to cancel the corresponding agreement by giving 90 days of written notice to the other side.
Anthropic will use the computing infrastructure to train new foundation models and operate its existing systems. Advanced AI models require vast computing power to analyze patterns across massive datasets, making server capacity the most valuable resource in modern software engineering.
Surging demand for supercomputing power
Anthropic is an artificial intelligence safety and research company best known for its Claude series of conversational assistants. Developing frontier language models requires enormous networks of specialized graphics chips capable of processing mathematical calculations simultaneously at high speeds.
Elon Musk has invested heavily in expanding data center facilities to compete in the global technology race. The Colossus supercomputing platform relies on tens of thousands of connected Nvidia chips, requiring specialized cooling and electrical infrastructure to maintain continuous performance.
Companies preparing for an initial public offering are required to outline major business commitments and spending obligations to prospective stock market investors. The filings show that computing expenses remain the single largest cost for developers competing at the cutting edge of artificial intelligence.
