Akamai's new contract with Anthropic has two very different clocks. The customer has agreed to spend approximately $11.6 billion on cloud capacity and related services over seven-year project terms. Akamai must build much of that capacity before the associated revenue appears. The company also granted Anthropic a warrant that could eventually represent about 5% of its shares. The central question for investors is how much cash the arrangement produces after equipment, operating costs and potential dilution, rather than how large the headline contract looks.
The September 24 announcement and the more detailed Form 8-K establish the commitment. They do not establish seven years of earned revenue today. Delivery and availability requirements apply, and the filing describes termination rights for breaches and certain material outages. The announced possibility of another $9 billion in business is an expansion opportunity, not part of the initial $11.6 billion commitment.
The purchase order is large; the income statement starts later
Akamai says the services will support Anthropic's CPU workloads on dedicated cloud infrastructure. That matters because this is a capacity project, not simply another customer using idle servers. Hardware procurement, colocation and commissioning determine when the service can start and when revenue can be recognized. Akamai expects no change to its 2026 revenue guidance from the new arrangement, according to its release. Its investor presentation projects service initiation in late second-quarter 2027, $150 million to $300 million of 2027 revenue, and a full contracted annual run rate only by the end of 2028. Those are management projections, subject to execution and change.
The contrast with Akamai's existing business shows the scale of the undertaking. Its second-quarter results put cloud infrastructure services revenue at $99 million for the quarter and companywide revenue at about $1.1 billion. Dividing $11.6 billion by seven years yields an illustrative average of roughly $1.66 billion a year, but that arithmetic is not the company's revenue schedule. The project starts later and ramps; treating the average as near-term sales would make the new business appear to arrive faster than Akamai has indicated. Nor should the earlier $2.8 billion of multiyear cloud commitments be casually added to an annual revenue figure: commitments are contract value across periods, while reported sales are earned in a period.
Hardware cash leaves before service cash arrives
Akamai estimates approximately $5.5 billion of capital expenditure connected with the $11.6 billion commitment. The company says it will increase 2026 capex by about $1.7 billion to secure components including memory. Its presentation allocates about $3.1 billion of additional project capex to 2027 and $0.7 billion to 2028. Those figures are estimates, but the order of events is economically important: a supplier can spend heavily before the contracted capacity produces much billed service.
The 8-K makes that supply chain unusually visible. It discloses a Lenovo hardware agreement and a Jabil build request authorizing about $1.7 billion of memory purchases. This is an authorized component purchase, not the contract's profit. Memory costs, delivery schedules, the usable life of equipment and the cost of running the facilities all affect the eventual return on invested capital. The filing says the customer obligation is subject to delivery and service availability conditions. A delay would therefore matter both to timing of revenue and to how long Akamai carries equipment and financing costs before utilization.
The bullish interpretation deserves space. A multiyear customer commitment can reduce the risk of building capacity with no buyer, and higher utilization could give Akamai's still small cloud infrastructure segment operating scale. Yet contracted sales minus projected capex is not a profit forecast. It omits electricity, colocation, networking, maintenance, depreciation, financing and taxes, and it ignores when each cash flow occurs. The company has projected strong cash conversion; reported cash flow as the build proceeds will test that claim.
The warrant pays for commitments that still need performance
Anthropic received a warrant to buy non-voting preferred shares convertible into about 7.7 million common shares, equivalent to up to roughly 5% of Akamai's common stock outstanding on an as-converted basis. The 8-K states that 40% of the warrant shares vest on the first payment under one project plan, subject to conditions. The remaining three equal tranches vest only as Anthropic commits an additional $3 billion of contractual value for each tranche. Exercise requires cash payment. Thus, the full 5% is neither already vested nor a free transfer of currently outstanding common shares.
This structure aligns part of the potential equity cost with an expanded commercial relationship, but it also complicates a simple revenue comparison. Akamai shareholders would have to judge the incremental value created by each additional commitment against both the new build requirements and the possible increase in shares. The initial warrant tranche also means a portion of that cost can arise before the optional expansion. An independent Reuters account corroborates the agreement and the contingent stake, while the filing supplies the precise conditions.
Deployment and cash conversion will settle the argument
The next useful evidence is operational. Investors can compare actual component spending and commissioning dates with Akamai's published timetable, then compare recognized cloud revenue, margins and cash generation with the stated ramp. Contract amendments, customer concentration disclosures and the amount of the warrant that actually vests would change the dilution side of the calculation. A smooth build with services starting on schedule could make the headline value more credible; cost overruns or a slower ramp could reduce returns even if the nominal commitment remains intact.
The contract is a substantial commercial win, but its economics are not determined by the $11.6 billion headline. The gap between cash invested now and services delivered later is the key test, and the warrant adds a second test of value per share. Both should be measured from future filings and realized performance, not inferred from the announcement alone.