Akamai’s $11.6B Anthropic Deal Is a $5.5B Bet on AI’s Boring Work
The AI gold rush is no longer just about Nvidia chips. Anthropic just committed $11.6 billion to Akamai for CPUs — and Akamai must spend $5.5 billion before the cash really rolls in.
Anthropic has just handed Akamai an $11.6 billion reason to stop being thought of as the company that helps websites load quickly.
The catch? Akamai now has to spend about $5.5 billion building the capacity to earn it. That is not a victory lap. That is a very expensive wager that the AI boom will keep chewing through ordinary computing power long after the headlines move on.
On September 24, Akamai announced a seven-year cloud-services commitment from Anthropic worth $11.6 billion. It can expand by another $9 billion, taking the potential relationship to roughly $20 billion. Akamai has also issued Anthropic a warrant that could eventually give the AI company an ownership position equivalent to as much as 5% of Akamai’s common stock.
This is the sort of deal founders and investors should study properly, not merely clap at on LinkedIn. It shows where the next pile of AI money is going, what infrastructure businesses are willing to risk to catch it, and why a big contract is not the same thing as easy money.
The $11.6 billion deal is not about shiny GPUs
The immediate temptation is to file this under “another massive AI infrastructure deal” and keep scrolling. That misses the useful bit.
Akamai says Anthropic will use its distributed cloud infrastructure and software to support growing CPU workloads. Not primarily GPU workloads. CPUs are the less glamorous workhorses of computing: the machinery that runs application logic, manages requests, coordinates systems, calls tools, handles data movement and executes plenty of the practical work surrounding an AI model.
That distinction matters because the commercial story of AI is moving beyond training a monster model once in a while. The money is increasingly in serving useful AI to real users, repeatedly, reliably and quickly.
A model that writes a clever paragraph in a demo is one thing. A model that runs agents, uses tools, queries systems, manages workflows and responds at scale is a different beast. That kind of product creates a lot of surrounding compute activity. Anthropic’s decision to commit this much to Akamai is a concrete signal that the boring plumbing around AI may become just as strategically valuable as the eye-watering accelerator chips everyone talks about at dinner.
That is not a prediction that GPUs suddenly become irrelevant. Far from it. It is a reminder that serious AI businesses need a whole factory, not one expensive machine.
Akamai has signed a giant cheque to itself first
Here is the part people should read twice: Akamai estimates that the capital expenditure tied to the $11.6 billion commitment will be about $5.5 billion.
It expects to increase 2026 capital expenditure by roughly $1.7 billion to secure and pre-purchase critical supply-chain components, including memory. Its investor presentation lays out another roughly $3.1 billion of spending in 2027 and about $700 million in 2028.
Meanwhile, Akamai expects no impact on its 2026 revenue guidance. Service is expected to begin in late second-quarter 2027, with revenue ramping in the second half of 2027. The company expects to reach the full contracted revenue run-rate only by the end of 2028, then recognise roughly $1.7 billion annually over the remainder of the contract term.
That is the real story.
Akamai is not receiving an $11.6 billion pile of cash next week and buying a boat. It is committing capital now, ordering scarce components now, building capacity now and trusting that it can deliver the infrastructure Anthropic needs on time and at a decent margin.
That is how infrastructure fortunes are made — and occasionally blown up.
The agreement is subject to delivery and service-availability requirements. In plain English: the headline value is meaningful, but execution still matters. Akamai has to build, install, operate and deliver. The contractual commitment gives it valuable revenue visibility, but no sensible operator confuses a seven-year commitment with seven years of frictionless cash collection.
Why Anthropic got the better deal than the headline suggests
Akamai did not simply win a customer. It gave Anthropic a reason to help make the relationship bigger.
The warrant covers non-voting convertible Series B preferred stock equal to 7.7 million Akamai common shares on an as-converted basis, or up to about 5% of Akamai’s outstanding common stock. The exercise price is $111.33 per share.
About 2% is expected to vest in connection with the current $11.6 billion commitment. The remaining roughly 3% vests if Anthropic expands the relationship by up to another $9 billion. Each additional $3 billion in cloud-services purchases can unlock about another 1% of Akamai’s common stock.
That is clever structuring.
Anthropic gets capacity and a potential financial upside if it helps turn Akamai into a more valuable AI infrastructure business. Akamai gets a marquee customer and aligns that customer’s incentives with expansion. Neither side is pretending this is a standard vendor invoice.
For Akamai, which has spent decades building a globally distributed network known for delivery and security, this is an attempt to convert old infrastructure muscle into a new growth engine. The company says its platform spans thousands of points of presence. In the AI era, proximity, reliability and the ability to distribute workloads can be commercially powerful — especially when customers need to serve products at scale rather than merely run a research lab.
But let’s not get carried away. Equity sweeteners are not free. If the relationship expands, the warrant can dilute existing Akamai shareholders. That might be a terrific trade if the additional business is profitable and durable. It is less terrific if investors celebrate the revenue headline while ignoring the capital intensity and dilution required to produce it.
The overlooked angle: this is a test of AI demand quality
The fashionable AI argument has been about capability: which model is smartest, which chatbot is growing fastest, which chip has the longest waiting list.
This deal asks a tougher question: will end-user demand become deep enough, persistent enough and monetisable enough to justify billions in dedicated compute commitments?
Anthropic has made a seven-year commitment. That is serious. But it also places real pressure on the company to turn usage into durable economics. AI infrastructure commitments do not disappear because a product manager changes their mind, a competitor launches a better model, or enterprise customers decide their pilot project was a bit of theatre.
The second-order consequence is that AI companies are becoming infrastructure allocators as much as software companies. They need to decide where capacity sits, what workloads run where, how much they lock in and how they avoid being trapped by yesterday’s technical assumptions.
Akamai’s deal is also a warning for founders building thin layers on top of foundation models. If the big AI labs are willing to make multibillion-dollar infrastructure bets around agents and production workloads, they are not planning to remain simple model providers. They are building industrial-scale operating systems for work.
If your entire business can be replicated by a model provider adding one more button, you have a problem. If your business owns customer workflow, proprietary data, distribution, trust and a painful operational job that people will pay to have done, you have a chance.
That has always been true. AI has just made the penalty for ignoring it much harsher.
The contrarian view: bigger AI contracts do not automatically mean better AI businesses
There is a nasty habit in markets of treating gigantic commitments as proof that everyone involved has already won.
No. It proves that everyone involved is prepared to take a big position.
Akamai is betting $5.5 billion of capital expenditure that Anthropic’s demand arrives as expected and that the company can serve it efficiently. Anthropic is betting that its products, pricing and market position justify long-duration infrastructure commitments. Investors are betting that the contract economics hold up after depreciation, financing, operations and the inevitable surprises that come with scaling physical infrastructure.
Those can all be good bets. They are still bets.
The useful lesson is not to become cynical about every large contract. It is to ask the boring questions before you celebrate the exciting number.
When does revenue actually begin? How much cash must be spent before it arrives? Is the contract conditional? Who bears supply-chain risk? What does the customer get in exchange for warrants or discounts? What happens if demand arrives slower than expected? And does the provider have other customers, or has it quietly become dependent on one enormous whale?
That is operator thinking. It is also how you avoid becoming the bloke who confuses a signed press release with profit.
What this means for you
If you are a founder, do this tomorrow: make a one-page dependency map of your business. List every platform, model provider, cloud provider, distributor and customer that can materially change your economics. Then write down what you own that they cannot easily take away.
If the answer is “our prompt engineering,” stop lying to yourself and build something sturdier.
If you are an operator buying AI infrastructure, negotiate for optionality before you need it. Locking in capacity can be smart. Locking yourself into one technical path because it looked clever during a fundraising round is not. Separate what must be committed from what should remain flexible.
If you are an investor, stop valuing AI companies solely on growth stories and model demos. Follow the capital commitments. They reveal what management actually believes it must build, what future revenue it needs to justify, and how much execution risk sits behind the glossy headline.
And if you run an old-school infrastructure business, take note. You do not need to invent the next Claude to benefit from AI. But you do need a genuine capability the AI boom cannot run without — and the guts to invest before the revenue is visible.
Akamai has shown both sides of that equation. Anthropic brought an $11.6 billion commitment. Akamai brought a $5.5 billion bill, years of execution risk and a willingness to bet that AI’s boring work will become very valuable indeed.
That is where the real business is: not in the hype, but in the machinery that survives after it.