Anthropic will spend $11.6 billion over seven years on Akamai’s cloud infrastructure, according to an announcement from Akamai on September 25. This marks the largest deal in Akamai’s history and could grow to roughly $20 billion if Anthropic expands its spending commitment. The figure shows that major AI labs require more than a single source of compute capacity, rather than relying exclusively on a handful of familiar providers.
According to TechCrunch, the deal is not an unconditional commitment. Akamai’s securities filings state that implementation depends on its ability to satisfy delivery requirements and service readiness. Both parties also retain the right to terminate the contract under certain circumstances. This detail is worth keeping in mind when looking at the $11.6 billion figure, as it is distinct from immediately recognized revenue.
Akamai Must Invest Before Seeing Substantial Revenue
Akamai stated that it does not expect any revenue from the agreement this year. The company anticipates a revenue contribution of $150 million to $300 million in 2027, beginning in the second half of the year, before gradually scaling to a run-rate of approximately $1.7 billion annually by late 2028. To secure enough capacity to serve Anthropic, Akamai expects to spend roughly $5.5 billion and add $1.7 billion in capital expenditures this year to pre-purchase components such as memory.
In other words, this is not a software contract with high profit margins from day one. Akamai must expand its infrastructure first, only recovering capital as Anthropic ramps up usage. This explains why the scale of the deal carries significant weight for Akamai while remaining tied to supply capacity conditions.
An unusual aspect lies in the financial structure. Akamai issued warrants to Anthropic for non-voting preferred stock convertible into 7.7 million shares of common stock, representing up to approximately 5% of its outstanding shares. About 2% will unlock when Anthropic makes its initial payment; the remainder is tied to incremental spending. Every additional $3 billion committed unlocks roughly 1% more.
This structure flips the familiar model where a chipmaker or cloud provider invests in an AI lab and then sells infrastructure back to that same customer. In this case, the provider grants the customer potential equity ownership. TechCrunch noted that AMD previously used a similar warrant mechanism in a chip agreement with OpenAI.
AI Demand Extends Beyond GPUs
The article described this as a bet on CPUs—general-purpose chips that typically handle tasks such as running code and browsing the web. Akamai did not disclose which specific workloads Anthropic will allocate this capacity toward. However, CPU demand could surge as AI agents coordinate multi-step workflows, execute tools, process background tasks, and handle massive query volumes. Such tasks do not always require top-tier GPUs.
This does not mean CPUs are replacing GPUs for large-scale model training or inference. GPUs remain an essential component of many AI workloads. However, the Akamai agreement illustrates that the cost of running AI agents encompasses the infrastructure layer surrounding the models: general-purpose servers, networking, memory, and high service availability.
Anthropic previously secured investment or infrastructure agreements with Amazon, Google, Microsoft, and AMD. Akamai’s deal is therefore not an indication that Anthropic is replacing its primary partners, but rather the addition of another piece to an expanding infrastructure portfolio. For Akamai, this represents both a major growth opportunity and a gamble on capacity buildout timelines and actual usage rates.
The key variable to watch in the coming quarters is not just the nominal $11.6 billion headline figure. More critical is whether Akamai can deliver capacity on schedule, whether Anthropic utilizes its committed levels, and how these infrastructure costs ultimately impact end-user pricing for AI products.