Inside Anthropic’s $11.6 Billion Akamai Compute Deal

Anthropic has committed to spending approximately $11.6 billion over seven years on dedicated cloud computing capacity and managed support from Akamai, marking a major expansion of the infrastructure behind its Claude AI services.

The companies entered the expanded project plans on September 18, 2026, and Akamai announced the agreement on September 24. It is not an upfront payment or a simple purchase of computer chips. The total represents contracted spending tied to Akamai delivering capacity and meeting service-availability requirements over the life of the arrangement.

What the agreement provides

Akamai will build and operate distributed cloud infrastructure designed primarily for Anthropic’s growing CPU workloads. Its network extends across numerous locations, allowing computing, storage and software services to be placed closer to users, business data and the applications that AI systems need to access.

The initial commitment could eventually grow by another $9 billion, taking the potential relationship to about $20 billion. Akamai estimates that it will spend approximately $5.5 billion in capital expenditures to support the currently contracted portion, including about $1.7 billion during 2026 to secure critical components such as memory.

Services are expected to begin during the second quarter of 2027 and expand through the following year. Akamai expects to reach the full contracted revenue rate, approximately $1.7 billion annually, by the end of 2028.

The agreement also includes a warrant giving Anthropic the right to buy non-voting convertible preferred shares representing up to roughly 5% of Akamai’s outstanding common stock. A portion connected to about 2% is expected to vest with the initial commitment, while the remainder depends on additional cloud purchases. That structure gives Anthropic a possible financial interest in the infrastructure provider as their commercial relationship grows.

Why an AI deal is focused on CPUs

Much of the AI infrastructure discussion has centered on graphics processing units, or GPUs. Their ability to perform many calculations in parallel makes them essential for training large models and generating responses. However, a production AI service requires much more than model computation.

CPUs handle general-purpose work surrounding the model, including routing requests, retrieving information, managing sessions, accessing databases, processing files, enforcing security controls and communicating with external services. They also operate the isolated environments in which coding agents can run commands, test software and inspect results.

These requirements grow as chatbots evolve into agents capable of completing multistep assignments. Instead of generating one answer, an agent may search for information, call several tools, execute code, examine files and repeatedly return to the model for its next decision. GPUs provide the intensive model calculations, while CPUs, memory, storage and networking keep the wider workflow moving.

Distributing some of this work can reduce the delays caused when every request must travel to a distant centralized data center. It can also prevent expensive accelerators from sitting idle while a workflow waits for a CPU-based task or external tool to finish.

A different front in the compute race

The Akamai agreement is only one part of Anthropic’s wider capacity strategy. In April 2026, the company announced a commitment of more than $100 billion over ten years for up to five gigawatts of additional Amazon computing capacity. Anthropic has also outlined a $50 billion investment in custom data centers in the United States.

Elsewhere in the industry, OpenAI and its partners have pursued hundreds of billions of dollars in planned infrastructure through Stargate. These projects illustrate how access to electricity, data centers, chips, memory and networking has become a competitive advantage alongside model quality.

The Akamai deal stands out because it concentrates attention on the supporting layer rather than only the largest accelerator clusters. It also gives Anthropic another infrastructure partner, potentially reducing dependence on a small number of hyperscale cloud platforms.

What users and competitors could see

For Claude users and developers, more capacity could mean fewer constraints during demand surges, greater reliability and faster performance for tool-heavy applications. A distributed network may be particularly useful for enterprise agents that need low-latency access to regional data and services.

The cost effect is less certain. Long-term commitments can lower per-unit expenses by giving a provider confidence to invest at scale. At the same time, the enormous capital required to build servers, secure memory and obtain data-center power creates fixed costs that AI companies must recover through subscriptions, usage charges or higher-value enterprise products.

For the cloud market, the agreement gives Akamai a chance to compete more directly for AI infrastructure spending traditionally captured by the largest providers. It may also encourage rivals to expand CPU-rich, memory-intensive and geographically distributed services. The larger message is that the AI capacity race is no longer only about obtaining the most GPUs. Increasingly, it is about assembling an entire computing system capable of serving millions of complex, continuously operating agents.