Anthropic Signs $11.6 Billion Akamai Cloud Deal
A seven-year CPU-heavy infrastructure contract gives Anthropic more capacity while tying Akamai’s future growth directly to Claude demand.
Anthropic has agreed to spend $11.6 billion on Akamai cloud services over seven years, in one of the largest disclosed infrastructure commitments by an AI model developer. The agreement expands an earlier relationship and is intended to support Anthropic’s growing CPU workloads for training, inference, and enterprise deployment.
A cloud deal with equity attached
Akamai issued Anthropic a warrant that could convert into roughly 7.7 million common shares, equivalent to about 5% of the company if the full relationship expands. Around 2% is linked to the initial commitment; the remaining portion depends on as much as $9 billion in additional purchases. Anthropic therefore does not currently own a 5% stake.
Akamai estimates that fulfilling the initial contract will require approximately $5.5 billion in capital expenditure. It plans to spend more on servers, memory, networking, and related components, with much of that investment expected next year. The agreement is unusually large for a company whose traditional business is content delivery and cybersecurity.
Why the infrastructure mix matters
The contract is notable because it centers on CPU capacity rather than a publicly disclosed dedicated GPU training cluster. As AI products become persistent services used by businesses, inference, orchestration, retrieval, and agent workloads can create substantial demand for general-purpose compute alongside accelerators.
For Anthropic, the deal diversifies its infrastructure base beyond hyperscalers and specialist data-center providers. For Akamai, it is a bet that its distributed cloud network can become a meaningful home for frontier-model workloads. The companies have not disclosed capacity by region, service-level commitments, margins, or whether Akamai will replace other suppliers.
The headline amount is a contractual commitment spread across seven years, not immediate revenue. Its significance lies in the scale of infrastructure being reserved before the next generation of agentic software fully arrives—and in the financial entanglement between a model maker and the provider building the capacity it needs.