Record-level investment spending in Akamai’s sights: initially, more than $3 billion slated for its next fiscal year.
Drumming up in the background is a contract of unprecedented scale as well. At least $11.6 billion over seven years, with the possibility to climb to $20.6 billion.
The client is none other than… Anthropic. It plans to leverage Akamai’s infrastructure for its CPU workloads.
Anthropic could own up to 5% of Akamai’s equity
Alongside this commitment, Anthropic gains the option to take a stake in Akamai. Akamai will issue, in tranches, warrants that will confer the right to preferred shares, with an exercise price of $111.33 (i.e., their value prior to the deal’s announcement).
A first tranche is released upon signing the initial firm commitment of $11.6 billion. It accounts for about 2% of the equity. Three more could follow, each conditioned on an additional $3 billion in orders. In the end, Anthropic could hold roughly 5% of Akamai.
This contract is expected to begin generating revenue from the second half of 2027. In the meantime, spending is required. Akamai anticipates a cash impact of $1.7 billion this year to “secure critical components, [notably] memory.” In total, it projects $5.5 billion in capex—more than a year’s worth of revenue (4.2 billion in 2025).
From neoclouds to hyperscalers, deals keep piling up
Anthropic is accustomed to deals of this scale. It does not always formalize them publicly. Take the “neoclouds” nScale and Lambda as examples. This summer, news broke of a $45 billion contract with the former over six years, for 460 MW of NVIDIA GPU capacity in Virginia. With the latter, it’s Texas, at a Hut 8 data center with 350 MW of available power. The arrangement is unusual. NVIDIA holds the lease and sublets the space to Lambda, which deploys its GPUs there.
There are contracts Anthropic discloses, but only selectively. The SpaceX agreement is one illustration. At the start of May, it was made official within the Colossus 1 data center (Tennessee; 300 MW). Two weeks later, Anthropic’s Chief Compute Officer spoke of an impending overflow at Colossus 2, located in the same geographic area. SpaceX ultimately revealed, in a pre-IPO document, a $45 billion contract over three years.
Anthropic has also decided to invest $50 billion in building its own data centers, in New York and Texas, with Fluidstack. It announced this at the end of 2025, nearly in parallel with a commitment to purchase around $30 billion of capacity on Azure.
A comprehensive setup with Google
Amazon is a supplier to Anthropic, but also an investor (up to $25 billion as of the latest tally). Anthropic plans to order from it, over 10 years, more than $100 billion of compute capacity.
The relationship with Google is of the same nature. And the business envisaged would be of a similar scale (around $200 billion). To back Anthropic’s deployment of its TPUs, Mountain View has put in place a mechanism involving banks, funds, and industrial players. The model, inspired by the financing of commercial aircraft, is designed to spare Google or Broadcom from having to book tens of billions of dollars of equipment on their own balance sheets. In broad strokes:
- Google sells the TPUs to Broadcom;
- Broadcom resells them to a financial vehicle funded by debt from Apollo and Blackstone;
- Anthropic leases the equipment from that financial vehicle.
Anthropic’s strategy with Google envisions a financing framework that keeps tens of billions of dollars off the balance sheets of Google or Broadcom while enabling rapid-scale deployment of TPU resources for Anthropic.
