Anthropic Akamai Deal: $11.6B Bet on CPUs for AI Agents
Anthropic just agreed to pay $11.6 billion for chips that aren't even the flashy kind. The Anthropic Akamai deal is about CPUs, the plain processors inside your laptop. It isn't about the GPUs that train AI models. And that choice tells you a lot about where AI is headed next.
Here's the short version. Anthropic, the company behind Claude, signed a seven-year cloud contract with Akamai on September 25. It could grow to about $20 billion, and Akamai's stock jumped more than 15% on the news.
Why should you care? Because this is the kind of computer power the AI agents you use will run on. It also shows how much money AI firms are betting on your future use.
What happened in the Anthropic Akamai deal
Anthropic will spend $11.6 billion over seven years on Akamai Cloud. That's about ?1.1 lakh crore at today's rate of roughly ?96 to the dollar. The deal covers Anthropic's fast-growing CPU work, according to Akamai's press release.
There's room to grow, too. Anthropic can add up to $9 billion more, which would push the total near $20 billion. The spending starts in the second half of 2027.
Then there's the twist. Akamai handed Anthropic a stock warrant, which is a right to buy shares later at a fixed price. It covers up to 7.7 million shares, or about 5% of Akamai, at $111.33 each.
About 2% vests with this deal. The rest only kicks in if Anthropic spends more, at 1% for every extra $3 billion. It's the first time Akamai has given a customer a warrant like this.
This isn't a first date, either. The two firms signed a $1.8 billion deal back in May. Akamai CEO Tom Leighton said his team was "thrilled" that Anthropic picked them.
Investors were thrilled too. Akamai's shares rose more than 15% as traders bet on years of steady AI income. Akamai says the deal won't change its revenue forecast for this year.
Why AI agents need CPUs, not just GPUs
Most AI news is about GPUs. Those are chips built to do huge piles of math at once. You need them to train a model and to write its answers.
But an AI agent does more than write. An agent is an AI model that takes actions for you. It might run code, open a web page or fill in a form. Those steps run on regular CPUs, as TechCrunch reports.
Picture a simple job, like asking Claude to compare flight prices on three sites. The model plans the job on GPUs. But opening each site, reading it and running the numbers happens on CPUs, again and again.
So the more agent work Claude does, the more CPUs Anthropic needs. We saw this coming when we looked at how Opus 5.5 cut the cost of agentic AI. Cheaper agents mean more agents. More agents mean more CPUs.
Akamai is a smart fit here. It started as a network that sends web pages from servers close to you. That same spread-out setup can run agent tasks near the people who ask for them.
How it compares
Big as it is, this isn't Anthropic's largest bet. Here's how the Anthropic Akamai deal stacks up against its other recent compute deals:
| Deal | Size | When |
|---|---|---|
| Akamai (first deal) | $1.8 billion | May 2026 |
| Nscale | $45 billion | August 2026 |
| Akamai (this deal) | $11.6 billion, up to about $20 billion | September 2026 |
Add it all up and the numbers get wild. The Decoder counts about $517 billion in Anthropic compute deals in just 11 months. It also notes that CEO Dario Amodei has warned bad forecasts could sink the company.
For Akamai, the payoff is slow. It has to spend about $5.5 billion to build the capacity first. It expects no revenue from the deal this year. That rises to $150 million to $300 million in 2027, and a pace of about $1.7 billion a year by 2028.
Here's what that ramp looks like:
- 2026: no revenue, but about $1.7 billion in extra spending on parts and memory
- 2027: $150 million to $300 million, starting in the second half
- 2028: about $1.7 billion a year
What the Anthropic Akamai deal means for you
If you use Claude to write code or browse the web for you, this is good news. More CPUs should mean less waiting when lots of people use it at once. That's the whole point of the deal.
India has a huge base of developers building on AI tools. If you're one of them using Claude's API, more capacity should help from late 2027. That's when the new Akamai spending kicks in.
Don't expect cheaper plans soon, though. Someone has to pay back these billions, and it's usually the people who subscribe.
More agents also bring more risk. Our look at the OpenAI agent breach in Australia shows what can go wrong when agents touch real systems. More power for agents should come with more care.
The bottom line
I think this is a sensible deal. Anthropic is paying for the dull chips that agents lean on, not just the famous ones. The real risk is the sheer pile of promises, and 2027 will show whether the demand is really there.
Read more on techmash
- Opus 5.5 lowers the barrier for agentic AI
- Why Google's orbital AI chips could change compute costs
- The rise of agentic AI, explained
- OpenAI agent breach: what Australia's Medicare hack means
Comments
0