Anthropic signs Akamai for 11.6 billion (expandable to 20 billion) in exchange for up to 5% of the company

By Zendoric · September 25, 2026.

Akamai Technologies announced on September 24 an expanded agreement with Anthropic worth $11.6 billion in contractual commitment over seven years, according to Akamai's statement. The contract covers the supply of Akamai's distributed cloud infrastructure to support the AI startup's growing demand for CPU workloads — the general-purpose processor that runs inference and other computing tasks, as distinct from GPUs specialized in training models.

The deal includes an expansion clause: if the relationship grows, the total commitment could reach $20 billion, adding up to $9 billion on top of the $11.6 billion already signed. As part of the agreement, Akamai has issued Anthropic a warrant — a right to buy shares at a price set in advance at a future date — covering 7.7 million convertible Series B shares, equivalent to up to roughly 5% of Akamai's equity, with an exercise price of $111.33 per share.

The fine print on vesting (the gradual accrual of the right to exercise those shares) matters: around 2% of the equity vests now, tied to the $11.6 billion commitment announced today; the remaining 3% will be released as the relationship expands toward those additional $9 billion, at a rate of roughly 1% of equity for every extra $3 billion in cloud service purchases. It is, in essence, an incentive: the more Anthropic spends on Akamai, the more Akamai shares it ends up owning.

On the financial side, Akamai puts the capital expenditure (capex, the investment in physical infrastructure) associated with this contract at some $5.5 billion. The company does not expect any impact on its 2026 revenue forecast, but it is raising this year's capex by $1.7 billion to secure critical supply chain components in advance, memory included. The agreement adds to more than $2.8 billion in other multi-year cloud infrastructure service commitments Akamai had already announced this year with other customers. After the news broke, Akamai shares rose sharply in the market, as reported by Benzinga. "Anthropic is driving the AI revolution and we are delighted that it has chosen Akamai's capabilities to build and operate AI infrastructure at scale," said Tom Leighton, co-founder and chief executive of Akamai, in the statement.

Broadly speaking, Akamai has spent years reinventing itself: it was born as the big name in content delivery networks (CDNs, the technology that replicates websites and video across thousands of servers around the world so they load quickly near the user) and, with that business mature and under pressure from rivals such as Cloudflare, it has gradually repositioned itself as a provider of "distributed cloud" and security for AI applications. This contract with Anthropic is the biggest proof to date that the bet has a buyer.

It is worth placing the agreement within the infrastructure map Anthropic has been weaving since 2025: training compute with Amazon's Trainium chips and Google Cloud's TPUs, plus the commitment of up to $30 billion on Azure announced in November 2025 alongside the joint investment by Microsoft and Nvidia of up to $15 billion in Anthropic itself. Akamai contributes a different piece: this is not a supercomputer race to train models, but a network of thousands of points of presence spread across the planet to serve inference and agentic applications close to the end user, with low latency. It is infrastructure for the phase in which models are already trained and hundreds of millions of simultaneous requests have to be handled.

The warrant mechanism is not an isolated novelty of this contract either: it is the same one AMD used with OpenAI in 2025, when it handed the lab a right over up to 10% of its equity tied to a multi-year chip order, and it recurs with variations across much of the major AI infrastructure contracts signed over the past year. The provider lands an anchor customer that guarantees future revenue; the customer obtains, in addition to the infrastructure, a slice of the stock market gains its own spending may generate for the provider. It is an elegant way to align incentives, but also a mechanism that makes a growing share of the sector's market value depend on those spending commitments — mostly promises stretching out over several years — actually being fulfilled.

Our read: in the short term, deals of this kind are fertile ground for legitimate suspicion. When one company announces it will buy $11.6 billion of services from another in exchange for taking a stake in it, the market revalues both companies on the basis of a promise of future spending, not of revenue already materialized; it is the same "circular financing" pattern that is beginning to unsettle analysts as it repeats among Nvidia, OpenAI, AMD, CoreWeave, Oracle and company. If AI demand slows, these multi-year commitments leave providers with capex already committed and capacity that may go spare.

But there is one detail in the statement itself that points to something more structural than financial engineering: Akamai is front-loading capex specifically to "secure" memory, one of the scarcest components in the semiconductor supply chain right now. That fits a thesis we have been advancing at Zendoric: AI's bottleneck is no longer so much cognitive — models keep improving — but physical: data centers, chips, memory and, now also, networks capable of serving that intelligence where the user lives. Whoever controls that material "plumbing" accumulates as much market power as whoever trains the smartest model. This contract is another piece of that industrial map: the construction of the physical substrate that, if AI's long-term bet pays off, will make it possible for artificial intelligence to stop being a scarce, expensive resource and become a utility accessible to practically anyone. The price of that transition, in the short term, is paid in warrants, soaring capex and a mutual dependence between labs and providers that will be worth watching closely.