Akamai says Anthropic deal could reach $20 billion, with warrant terms tied to spend

Akamai said on Sept. 24 that it has signed a seven-year, $11.6 billion cloud commitment with Anthropic that could expand by another $9 billion, bringing the total potential value to about $20 billion. In Akamai’s announcement, the company said the agreement is meant to support Anthropic’s growing CPU workload demand on Akamai Cloud. TechCrunch reported that the contract is the largest in Akamai’s history and that its performance depends on delivery and service-availability conditions.

What changed

The immediate change is scale. Akamai is no longer describing Anthropic as a standard cloud customer; it is describing a multi-year commitment large enough to reshape its infrastructure planning and, potentially, its shareholder structure. The company said it has also issued Anthropic a warrant tied to the relationship, with vesting that can reach roughly 5% of Akamai’s common stock if the partnership expands as planned. A portion representing about 2% is expected to vest with the announced commitment.

That detail matters because it turns the deal into more than a simple services contract. The financial relationship now has a second layer: the more Anthropic spends, the more of the warrant can vest. In practical terms, that is an incentive system designed to keep both sides aligned over a long horizon, while also making the final economics dependent on execution rather than just signature value.

Why CPUs matter here

Akamai framed the arrangement as support for Anthropic’s accelerating CPU workload demand. The company did not spell out the exact workloads, but that wording is revealing. In AI infrastructure, CPUs often sit in the stack around model inference, orchestration, agent execution, networking and application logic, even when GPUs still dominate frontier training. That suggests the deal is not only about raw model training capacity; it is also about the broader runtime that lets AI systems operate in production.

That is an editorial inference, not a disclosed technical map. Still, it helps explain why this is interesting beyond the headline number. A large commitment for CPU-heavy AI work points to the less glamorous but increasingly important layer of the AI market: the infrastructure used by applications and agents once the model itself is already built.

The economics are as important as the headline

Akamai said total capital expenditures related to the announced $11.6 billion commitment are estimated at about $5.5 billion, and that it expects no change to 2026 revenue guidance. The company also said 2026 capital expenditures will rise by about $1.7 billion to secure and pre-purchase components, including memory. That means the revenue story and the cash-spend story are not synchronized: Akamai is taking on buildout costs now for revenue that will, by its own account, arrive later.

TechCrunch, citing the company’s investor call, reported a more specific ramp: executives expect $150 million to $300 million in revenue in 2027, starting in the second half, with annualized revenue reaching about $1.7 billion by the end of 2028. If those projections hold, the deal would evolve from a backlog item into a major line of business. If they do not, the capital intensity becomes the story instead.

Why the warrant structure is unusual

The warrant is the most notable structural twist. In many circular AI arrangements, the cloud provider or chipmaker takes an ownership stake in the model company. Here, Akamai is giving Anthropic the potential to own part of Akamai, and only if Anthropic keeps expanding its spend. That reverses the usual direction of strategic equity and makes the supplier-customer relationship feel more like a long-duration partnership with milestones attached.

TechCrunch described the deal as a flip of the more common pattern in which infrastructure vendors invest directly in the labs that buy their products. That does not make the structure inherently good or bad, but it does make the incentives easier to read: Akamai wants committed usage; Anthropic gets a path to ownership if the relationship deepens. The trade-off is obvious too. The more the two companies’ economics depend on one another, the more important delivery quality, pricing discipline and service reliability become.

What readers should take from it

For investors, the key question is whether Akamai can turn a huge announced commitment into durable, high-quality revenue without overspending on capacity. For Anthropic, the obvious advantage is access to another large-scale infrastructure pool outside the hyperscaler center of gravity. For the broader AI market, the deal is another sign that CPU-heavy infrastructure is becoming strategically relevant as AI systems move from training demos into deployed products and agents.

One caveat is essential: the core figures come from Akamai’s own announcement, so they are vendor claims, not independent proof of execution or utilization. TechCrunch’s reporting adds useful context on conditionality and timing, but it does not remove the forward-looking nature of the deal. The real test is whether the contract’s milestones, service requirements and warrant vesting actually track through the next earnings cycle.

Watch Akamai’s future reports for service start dates, capital spending, the revenue ramp and warrant vesting. Those milestones will show whether the announced commitment is moving into delivery.