Akamai announced on 24 September 2026 that Anthropic has committed $11.6 billion over seven years for infrastructure, with an option to expand by a further $9 billion to a ceiling near $20 billion. Akamai stock rose more than 20%.
The headline number is not the interesting part. The interesting part is what Anthropic is buying.
This Is a CPU Deal
Every large AI infrastructure announcement of the last three years has been about GPUs. This one is not. Akamai's commitment is for CPU workloads on its distributed infrastructure.
That tells you something about where Anthropic's costs actually sit. Training needs GPUs. Serving a model needs GPUs. But everything around the model - orchestration, tool calls, retrieval, sandboxing, the actual running of agents - is ordinary compute, and at agent scale there is an enormous amount of it.
Anthropic ran roughly 950 agents for 21 hours on a single research sweep this week. Multiply that shape across a customer base and the non-GPU bill stops being a rounding error. An $11.6 billion CPU commitment is a company telling you that agents, not inference, are the thing it is scaling for.
The Warrant Is the Unusual Part
Anthropic receives a warrant for up to 5% of Akamai's common stock - roughly 7.7 million shares - at an exercise price of $111.33 per share.
- About 2% vested immediately on announcement of the $11.6 billion commitment
- The remaining 3% is tied to expansion milestones
- Roughly 1% vests per additional $3 billion of cloud services purchased
Read that structure carefully. The customer gets equity in the supplier, and gets more of it the more it spends. Anthropic is not simply buying compute - it is taking a position in the company building the compute, with the position growing as the commitment does.
That is a hedge. If Akamai's AI infrastructure business succeeds on the back of this deal, Anthropic owns a slice of the upside it created.
What It Costs Akamai To Say Yes
Akamai estimates roughly $5.5 billion of capital expenditure to service the $11.6 billion commitment, with about $1.7 billion of additional capex in 2026 alone, largely for supply chain and pre-purchasing memory.
Pre-purchasing memory is worth noting on its own. It is a company locking in components years ahead because it expects them to be scarce - which is a quiet signal about where the hardware market is going.
Tom Leighton, Akamai's chief executive: "Anthropic is advancing the AI revolution and we are thrilled they chose Akamai's capabilities for building and operating AI infrastructure at scale."
Why a CDN Company
Akamai's core asset is a distributed network - compute in many places rather than compute in a few enormous buildings. For training, that is useless. For agents, it is close to ideal: agent workloads are latency-sensitive, bursty, and want to run near the data and the user rather than in a single region.
If your bet is that the next two years are about agents doing work rather than models answering questions, distributed CPU capacity is exactly what you buy. This deal is that bet, written down with a number attached.
What It Means For You
- Your agent costs are not just token costs. The orchestration around the model has a bill, and the frontier labs are now sizing it in the billions. If you are budgeting agents on token price alone, you are budgeting one line of two.
- Capacity is being locked up years ahead. Seven-year commitments and pre-purchased memory mean the supply available to everyone else is being spoken for now.
- Watch for the same structure elsewhere. Compute-for-equity is a way for a capital-hungry lab to buy infrastructure without paying entirely in cash. Expect to see it again.
Sources