Anthropic agreed to spend $11.6bn over seven years on Akamai cloud infrastructure, with an option to add another $9bn and take the total to about $20bn. Akamai disclosed the agreement after the US market closed on Thursday and granted Anthropic a warrant for about 5% of its common stock. The contract builds on a $1.8bn deal the companies signed in May. For business buyers of AI capacity, the structure shows how compute suppliers now lock in long demand with equity-linked terms.
Terms of the $11.6bn commitment and warrant
Under the base commitment, Anthropic received a warrant for 7.7 million Akamai shares on an as-converted basis at an exercise price of $111.33. The instrument takes the form of non-voting convertible Series B preferred stock. About 2% vests against the $11.6bn commitment, while the remaining 3% vests only if the contract expands, at 1% for every $3bn of additional services. Bloomberg Lynn Doan reported that Akamai will supply access to central processing units (CPUs). Akamai shares rose as much as 17% in late trading to $129.60 after the disclosure, according to Bloomberg.
Akamai expects to spend about $5.5bn in capital to serve the contract, more than six times its capital spending in all of 2025, according to Bloomberg. It is adding about $1.7bn to 2026 capital spending to pre-buy supplies, including memory, with no change to 2026 revenue guidance. Chief executive and co-founder Tom Leighton said the contract should start in the second half of next year. He put revenue from Anthropic at $150m to $300m next year, rising to an annual run rate of about $1.7bn by 2028. He said this is the first time Akamai has granted a warrant in a customer cloud deal, describing it as a serious step that brings the companies together.
Akamai chief executive said Anthropic selected Akamai for building and operating AI infrastructure at scale, while Anthropic was not quoted in the announcement. The contract extends a run of large compute moves, including a $10bn deal with Volta in August. In September, The Information reported Anthropic as the customer behind Rum Group $13.7bn compute contract, and Anthropic also signed its first Australian data centre lease. Bloomberg reported additional chip supply from Google and SpaceX. Bloomberg also noted Wall Street concern about circular AI deals, where commercial contracts sit alongside cross-investments and obscure the level of underlying demand.
What this means for buyers of AI capacity
For companies running AI workloads, the deal widens the supplier set beyond the largest clouds. Akamai contracted resource is CPU access, which points to scaled inference and operating infrastructure rather than frontier training alone. A seven-year term with a ramp toward a $1.7bn annual run rate by 2028 gives large buyers a reference for multi-year capacity planning. Smaller firms will not sign such contracts, but they may benefit from wider regional availability and combined delivery-plus-compute supply. Larger buyers gain an additional bidder for long agreements, with more room to negotiate duration, expansion rights and supply assurance.
Before drawing procurement conclusions, buyers should separate the headline total from funded timing. Only $150m to $300m is expected next year, with the start in the second half of next year, and $9bn of the $20bn figure depends on an expansion decision. The warrant structure also ties supplier alignment to equity upside, with most of the 5% vesting only on expansion, which deserves legal and financial review. The circular-deal debate is a further check, since overlapping contracts and investments can flatter demand signals. Buyers should ask vendors about delivery milestones, hardware availability including memory, and what happens to pricing and priority if expansion is not triggered.
The marker to watch is whether Akamai books $150m to $300m from Anthropic next year and moves toward the $1.7bn run rate by 2028. A decision on the extra $9bn, visible through additional warrant vesting at 1% per $3bn, will show whether the $20bn outline becomes a firm program. Those two data points will indicate if CPU-based capacity from a delivery-network operator can hold a durable place in AI infrastructure supply.
