Anthropic pays for cloud with contracts and collects in shares: Akamai hands it up to 5% of its equity for $11.6 billion

By Zendoric · September 26, 2026.

Anthropic has committed to paying Akamai Technologies at least $11.6 billion over seven years for cloud infrastructure and software, according to an announcement by Akamai itself. The deal expands an earlier alliance between the two companies and responds, again according to Akamai, to Anthropic's growing demand for CPU workloads (the general-purpose processors that complement GPUs in training and, above all, in the day-to-day running of models). Akamai shares rose in after-hours trading following the news, as reported by The Wall Street Journal.

The contract includes an expansion clause: Anthropic could spend up to an additional $9 billion, which would take the total commitment to roughly $20 billion over seven years and make Dario Amodei's company one of Akamai's largest customers by contract value. That figure comes on top of more than $2.8 billion in other multiyear cloud infrastructure deals Akamai has announced with other customers so far this year alone, a sign that its cloud business is growing strongly beyond this particular contract.

What is unusual about the deal is not so much the figure as the quid pro quo. Akamai has issued Anthropic a warrant — an option granting the right, but not the obligation, to buy shares at a price set in advance — for 7.7 million common shares, up to 5% of Akamai's equity, at an exercise price of $111.33 per share. Around 2% of that potential stake vests immediately along with the $11.6 billion commitment; the remaining 3% unlocks only if Anthropic carries out the additional $9 billion in spending, at a rate of one extra percentage point for every $3 billion spent. To prevent an AI customer from ending up with a vote on the board of a critical infrastructure provider, the instrument has been structured as non-voting convertible preferred stock.

For Akamai, executing the contract will cost around $5.5 billion in capital expenditure (capex), according to its own figures, and the company says it does not expect the deal to change its 2026 revenue forecast. It does anticipate an increase of about $1.7 billion in this year's capital spending to secure and pre-purchase supply-chain components, memory included, ahead of building the infrastructure. It is a telling detail: it confirms that memory (the DRAM and HBM chips that power AI servers) has become the sector's new bottleneck, with suppliers buying ahead as prices escalate. Akamai co-founder and CEO Tom Leighton said the company is pleased that Anthropic has chosen its capacity to build and operate AI infrastructure at scale, and highlighted Akamai's experience serving large enterprises as a selling point for future secure AI applications.

As sector context, this kind of warrant handed by an infrastructure provider to an AI lab is not an Akamai invention: AMD struck a similar deal with OpenAI barely a year ago, and Nvidia and Oracle have woven equally intertwined relationships — compute contracts in exchange for equity stakes or cross purchase commitments — with several of the major labs. A pattern is taking hold: when a non-hyperscaler provider needs a revenue anchor to justify billions in capex, it offers equity; when an AI lab needs to diversify its compute supply beyond its two usual partners (Amazon and Google, in Anthropic's case), it accepts that equity as an added incentive on top of the contract.

Our reading is that this deal has to be read on two different levels. In terms of real demand, it is good news: Anthropic needs more compute to serve a product that is growing, and Akamai — a company born as a content delivery network (CDN) that has spent years reinventing itself as a cloud provider after buying Linode — finds in AI the growth lever it had long been looking for. Financially, however, this kind of circular structure (the customer receives shares in the provider it is paying, and the price of those shares depends in part on the customer continuing to pay) deserves scrutiny: it is not clear how much of the stock-market gains these announcements celebrate reflects end demand for AI products and how much comes down to the mechanics of the contract itself. It is exactly the kind of financial plumbing that can sustain an industry's narrative for a while without yet translating into proportionate revenue for anyone.

In the long run, though, this is simply the cost of building the physical infrastructure needed for frontier models to keep scaling: more distributed compute capacity, more providers competing, more pressure to bring down the cost of access. If the underlying thesis about AI — that it can accelerate cures for diseases and generate abundance — holds true, it necessarily runs through transactions like this one, however opaque or financially ingenious they may look today. The short-term risk is that the bill for that build-out — in capex, in shareholder dilution, in strained memory supply chains — gets paid before real demand fully backs it up.