Anthropic to Pay Akamai $11.6 Billion for CPU Infrastructure for AI Agents

Akamai revealed on Thursday, September 24, 2026, that Anthropic will pay $11.6 billion over seven years for the company's cloud capacity, according to a TechCrunch report. It's the largest contract in Akamai's history and more than six times the size of a previous deal between the two companies, worth $1.8 billion, reported by Bloomberg in May of this year.

The commitment isn't ironclad: according to Akamai's filing with regulators, it depends on the company meeting delivery deadlines and service availability targets, and either party can terminate the agreement under certain conditions. Even so, the volume gives a sense of how much Anthropic keeps devouring computing capacity to sustain its models and products.

Why CPU, not GPU

The detail that sets this contract apart from the wave of billion-dollar GPU investments is the type of chip involved. Akamai hasn't detailed what Anthropic will use the capacity for, but TechCrunch points out that demand for CPUs (the general-purpose chips that run code and browse the web) has been growing as AI agents take on more tasks.

This matters for those building agent systems: training and serving inference for large models is a GPU game, but orchestrating an agent that runs tools, spins up code sandboxes, browses pages, and calls APIs is, in practice, a CPU and I/O problem. An agent that opens a headless browser, runs a Python script, or parses an HTML page consumes general-purpose computing, not tensor cores. If the agent adoption curve keeps rising, teams that today only plan for GPU inference costs may be underestimating the real spend: the execution layer around the model can weigh as much as the model itself.

Akamai historically comes from the CDN and content delivery world, with a globally distributed edge network. Repositioning that infrastructure for AI agent workloads is consistent with what the company itself has been signaling to the market: computing closer to the end user, useful both for serving content and for running the light, frequent tasks that characterize an agent in production.

How the financial return works on Akamai's side

Akamai won't book any revenue from the deal this year. According to executives, on an investor call on Thursday, the expectation is $150 million to $300 million in revenue in 2027 (starting in the second half), reaching an annual run rate of about $1.7 billion by the end of 2028.

To build that capacity, Akamai expects to spend around $5.5 billion, and is adding $1.7 billion to this year's capex just to buy components like memory in advance, likely to protect against supply bottlenecks.

The warrant that flips the logic of circular deals

The most unusual point of the deal is its financial structure. As part of the contract, Akamai issued Anthropic a warrant, that is, the right to buy shares at a fixed price, covering non-voting preferred shares convertible into 7.7 million common shares, up to about 5% of Akamai's outstanding stock, at $111.33 per share.

About 2% of that total becomes exercisable as soon as Anthropic makes the first payment on the contract. The rest is tied to spending more: each additional $3 billion committed to Akamai's cloud services unlocks roughly another 1%. In practice, this means the deal could grow by up to $9 billion more, reaching about $20 billion in total, if Anthropic keeps scaling its consumption.

This is the first time Akamai has tied a warrant to a cloud contract, and TechCrunch notes that this flips the more common pattern in so-called AI "circular deals," in which suppliers (chipmakers and cloud providers) invest directly in the AI labs that buy their products. Here it's the opposite: the infrastructure supplier is the one handing a potential stake to the customer, growing as the customer spends more. AMD used a similar structure with OpenAI last year, tying warrants to chip purchase milestones.

Anthropic is no stranger to this kind of arrangement: Amazon, Google, Microsoft, and AMD itself have already invested or committed to invest in the company while also selling it chips or cloud capacity. Last December, CEO Dario Amodei told the New York Times that Anthropic doesn't take part in such deals "at the same scale as some other players," a caveat that carries extra weight precisely in a contract of this size.

The market reaction was immediate: Akamai's shares rose as much as 17% in after-hours trading on Thursday, according to the Wall Street Journal.

What remains open for those building AI infrastructure

For engineering teams in Brazil that depend on American cloud to train, fine-tune, or serve AI agents, the episode reinforces two practical points. First, agent infrastructure economics isn't only about GPU: if CPU and memory become the bottleneck (Akamai itself is buying memory ahead of time as a precaution), the cost of running agent orchestration at scale can rise independently of the model's inference price. It's worth reviewing cloud contracts and quotes with an eye on the agent's tool execution layer, not just the model provider.

Second, the proliferation of warrants and cross-holdings among cloud providers, chipmakers, and AI labs concentrates even more pricing power in this handful of American companies. This doesn't change the code a dev writes tomorrow, but it explains why price volatility and AI capacity availability tend to stay high: when customer and supplier become partners in each other, the incentive to consume grows, and whoever pays the bill at the end of the chain is whoever uses the API.

The contract's conditional termination clause, tied to Akamai's delivery targets, is also a reminder that even billion-dollar commitments like this one are no guarantee of capacity: they depend on operational execution that's still to come over the next seven years.