Akamai Anthropic Cloud Deal: $11.6 Billion, Seven Years, 5% Stake Warrant

Akamai Technologies has agreed to supply Anthropic with $11.6 billion of cloud capacity across seven years and will issue the AI developer a warrant covering up to about 5% of Akamai's common stock. The Akamai Anthropic cloud deal took effect September 24, 2026, and can expand by a further $9 billion in purchases, lifting the total potential commitment to roughly $20.6 billion. Akamai shares rose about 22% in extended trading after the disclosure, which ranks as the company's largest AI customer commitment.

What separates this contract from a conventional capacity sale is the payment structure. Anthropic receives non-voting convertible Series B Preferred Stock equivalent to roughly 7.7 million common shares, carrying an exercise price of $111.33 per share. About 2% of Akamai's common stock vests against the initial commitment. The remaining 3% vests one percentage point at a time for every $3 billion Anthropic spends beyond the base agreement.

That ladder turns a procurement plan into an ownership schedule. Each additional dollar of cloud spending raises Anthropic's share of the company selling it. Akamai gains demand visibility it can build capacity against, and Anthropic converts a multi-year operating expense into an equity position whose value depends on Akamai's share price clearing the fixed strike.

The Numbers Behind the Commitment

Stripping the agreement to its components makes the cash and equity flows easier to follow.

Capital Spent Years Before Revenue Arrives

The timing is where the risk sits. Akamai expects related capital expenditures of about $5.5 billion, and a large share of that is inventory bought ahead of need: roughly $1.7 billion of memory components purchased to secure supply. Revenue from the agreement does not begin until late 2027, then ramps toward approximately $1.7 billion a year by late 2028.

Memory accounts for the visible portion of the near-term spending. The $1.7 billion of pre-purchased memory components lines up with the $1.7 billion increase to Akamai's 2026 capital expenditure plans, which indicates the current-year increase is largely inventory positioned ahead of the contract. Buying components early hedges against supply constraints and price movement, and it fixes Akamai's cost basis before demand converts into revenue.

Read the capital figure and the revenue schedule together and the shape of the bet becomes clear. Akamai commits billions in capacity through 2026 and 2027 against a contract whose first meaningful revenue arrives a year later. The $11.6 billion headline is spread across seven years, so the annual contribution is a fraction of the total. Akamai has confirmed the agreement leaves its 2026 revenue guidance unchanged.

Divide the base commitment by its term and the contract averages about $1.66 billion a year, which sits close to the roughly $1.7 billion annual run-rate Akamai expects by late 2028. That near-match indicates the spending is back-loaded: the early years contribute little, and the agreement approaches full run-rate only once the build-out is finished. The schedule explains why Akamai framed the announcement around capital expenditure rather than bookings.

The warrant's cash value to Akamai is small next to the contract. Exercising all 7.7 million shares at $111.33 would bring in roughly $857 million, far below the $11.6 billion Anthropic has committed to spend. The warrant is about alignment rather than financing: it gives Anthropic a reason to route more traffic to Akamai and gives Akamai a shareholder whose interests track its own capacity build-out.

The vesting ladder is calibrated to the contract's ceiling. Anthropic vests the full 5% only by spending the entire additional $9 billion, so the equity transfer scales one for one with the demand Akamai is underwriting. Partial expansion produces partial dilution, and the dilution arrives alongside the revenue that triggers it.

The 22% share move indicates the market treated the agreement as an option on future inference demand rather than a near-term earnings event. That reading is defensible: the contract adds a long-dated revenue stream and a customer with equity incentives to keep buying.

Why Anthropic Is Buying CPUs

Anthropic's commitment covers CPU workloads distributed across Akamai Cloud and a network spanning thousands of points of presence. AI labs usually sign their largest infrastructure agreements for GPU clusters dedicated to training, which makes the emphasis here unusual.

A CPU-heavy contract points at the serving side of the stack: inference requests, routing, and the orchestration layer that coordinates agents making repeated calls to models. Akamai describes the driver as accelerating CPU workload growth at scale. Those jobs reward proximity to users and predictable latency more than raw accelerator density, which suits a network assembled from many edge locations rather than a small number of hyperscale campuses.

Agentic systems compound that demand. A single user task handled by an agent can generate dozens of sequential model calls, each requiring routing, state handling and tool execution on conventional processors. The ratio of CPU work to accelerator work rises with that pattern, which is the growth Akamai is underwriting.

The competitive implication is that Akamai is not contesting the training capacity that the largest cloud providers fight over. It is competing for inference and orchestration, where centralized regions are not automatically the best fit. That is a narrower market, and it is one where Akamai's existing footprint does work that would otherwise require a greenfield build.

Who Carries Which Risk

For Akamai shareholders, the dilution is capped near 5% and is earned through customer spending rather than granted at signing, which makes it closer to a volume rebate paid in stock than a transfer of ownership. The counterargument is that a customer holding 5% of its supplier carries influence a plain contract does not convey, which is likely why the shares are non-voting.

For Anthropic, the exposure is concentration. Committing $11.6 billion over seven years to one provider ties a large share of its compute roadmap to that provider's build-out schedule. The equity stake offsets part of that: if Akamai executes and the shares appreciate, the warrant's value covers some of the cost of the capacity.

The Akamai Anthropic cloud deal also raises a capacity question for Akamai's existing customers. Pre-purchasing $1.7 billion of memory and building distributed compute for a single anchor tenant shifts the company's investment toward AI serving workloads. If that capacity is fungible, the build-out widens what Akamai can sell to other enterprises; if it is dedicated, the company's growth story rests more heavily on one customer's roadmap.

The unresolved variable is demand durability. Akamai is spending $5.5 billion against a contract whose expansion is optional and whose warrant tranches vest only if Anthropic keeps purchasing. If inference demand keeps growing, the pre-purchased memory and the distributed build-out look inexpensive. If growth flattens, Akamai holds underused capacity and a larger share count, with the first revenue still a year away.

Why this matters

The Akamai Anthropic cloud deal shows that equity has become a working lever in AI infrastructure negotiations rather than an exception. Buyers of compute now negotiate ownership alongside price and capacity, and suppliers fund capacity years before it earns anything. For decision-makers weighing similar arrangements, the number that matters is how much capital the supplier commits before revenue starts, and what becomes of that capital if the buyer's demand curve bends. Akamai's answer is a $5.5 billion bet that inference traffic keeps growing through late 2027.

Related Articles

• Anthropic Compute Commitments Near $517 Billion as Its Pacing Message Wears Thin
• Anthropic 1 Gigawatt Data Center Lease Talks Shift Compute Risk to Private Capital
• Amazon Increases Anthropic Investment to $13 Billion in Massive Cloud Expansion

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.