Akamai Technologies has signed an $11.6 billion, seven-year cloud infrastructure agreement with Anthropic, the largest deal in the company’s nearly 30-year history, and its shares jumped as much as 20 percent in after-hours trading when the deal was announced Thursday. The contract, confirmed by Reuters, covers CPU workloads on Akamai Cloud’s distributed infrastructure, not the GPU-heavy training runs that dominate AI spending headlines, and it includes an option for up to $9 billion in additional commitments that would take the total potential value to roughly $20 billion.
As part of the arrangement, Akamai issued Anthropic a warrant for up to about 5 percent of its common stock, with roughly 2 percent vesting immediately and the remainder vesting in blocks tied to additional spending. The warrant is struck at $111.33 per share, and its structure gives the AI lab a direct equity stake that grows if the relationship expands.
What the money buys
Anthropic will use Akamai Cloud to run the CPU-side work that keeps AI products running: serving applications, orchestrating agent tasks, processing data and handling the unglamorous plumbing around model inference. Revenue from the contract is expected to begin in the second half of 2027 and reach an annualized run rate of about $1.7 billion by the end of 2028, according to company guidance. Akamai plans roughly $5.5 billion in capital expenditures through 2028 to build the required capacity, including a $1.7 billion increase in 2026 spending to pre-purchase critical supply chain components such as memory.
The deal lifts Akamai’s 2026 year-to-date signed contract value to approximately $14.4 billion, adding to more than $2.8 billion in multi-year cloud infrastructure commitments announced earlier in the year. For a company that built its reputation on content delivery and edge security rather than hyperscale AI compute, the contract marks a decisive pivot toward the AI infrastructure market. Akamai’s edge network spans thousands of points of presence, and the company is betting that AI workloads increasingly want compute close to users rather than concentrated in a handful of hyperscale regions.
Why CPU workloads are suddenly valuable
The deal lands in the middle of a repricing of CPU demand across the industry. Meta’s Muse personal AI agent, launched September 8, has pushed chip stocks sharply higher on the insight that persistent AI agents consume far more ordinary compute than chatbots do. Each Muse user gets a dedicated cloud virtual machine that keeps working after the app closes, and industry estimates put CPU-to-GPU ratios for agent workloads between 4:1 and 40:1, the inverse of the training-era assumption that AI spending flows mostly to accelerators.
That shift has already moved markets. AMD crossed a $1 trillion market capitalization for the first time last week, Intel gained 12 percent and Arm Holdings climbed 17 percent in a single session as investors bet on CPU demand from agentic AI. Akamai’s contract shows the same logic applying to cloud infrastructure: agents need places to run, and those places are provisioned in vCPUs, RAM and storage rather than GPU hours alone.
Analysts at JPMorgan called the option for an additional $9 billion a path of low resistance to exercise, noting that Anthropic’s compute needs have grown faster than its existing cloud commitments can absorb. The company’s spending across AWS, Google Cloud and other providers has scaled steeply through 2026, and diversifying across a distributed edge network gives it capacity that the big three hyperscalers price differently.
Context: Anthropic’s infrastructure spending
The Akamai deal is one piece of a much larger infrastructure buildout. Anthropic has been one of the most aggressive buyers of compute among AI labs, and the industry’s memory and equipment supply chains have reorganized around AI demand this year. Chip equipment component lead times have reportedly doubled to as long as 40 months, and memory makers have shifted capacity toward high-bandwidth memory for accelerators, tightening conventional supply and raising prices across the board.
The spending race has also drawn scrutiny. Oracle this week sent a force majeure notice to the developer of its Project Jupiter data center campus in New Mexico, seeking payment relief if the facility misses its 2028 online target, a signal that even the largest AI infrastructure projects face execution risk. Against that backdrop, Akamai’s contract is notable for its structure: the revenue only starts in 2027, giving the company time to build before the bill comes due.
| Term | Detail |
|---|---|
| Contract value | $11.6 billion over seven years |
| Expansion option | Up to $9 billion additional, roughly $20 billion total |
| Workload type | CPU workloads on Akamai Cloud distributed infrastructure |
| Revenue start | Second half of 2027 |
| Run rate by end of 2028 | About $1.7 billion annualized |
| Capex through 2028 | Roughly $5.5 billion |
| Warrant | Up to about 5% of Akamai shares at $111.33 |
Market reaction and what to watch
Akamai shares rose roughly 20 percent in after-hours trading on the announcement and extended gains in premarket trade on Friday, with Morningstar noting the deal reveals an edge the company holds over traditional cloud providers in distributed, edge-proximate compute. The stock had been one of the quieter large-cap tech names this year, and the contract repriced it in a single session.
Watch three things from here. First, whether Anthropic exercises the $9 billion option, which would make this one of the largest cloud contracts ever signed outside the hyperscalers themselves. Second, whether other AI labs follow with similar CPU-side deals, since OpenAI, Google and others face the same agent-workload economics. Third, how quickly Akamai can execute the buildout, because the contract’s value depends on capacity arriving on schedule in a supply chain where memory and equipment components are increasingly hard to secure.