CAMBRIDGE, Mass. — Akamai Technologies (NASDAQ: AKAM) shares jumped more than 20 percent Wednesday, adding roughly $3 billion to the company’s market capitalization in a single session. The catalyst was not a new product or an earnings report. It was a seven-year contract.
The company disclosed a $11.6 billion commitment from Anthropic, the largest single AI cloud services deal announced by any company in 2026, with a contractual option to expand the total to $20 billion. The underlying agreement took shape in stages: a master services arrangement signed as early as May 5, 2026, followed by the two project plans that crystallized the $11.6 billion commitment on September 18. Six days later, Akamai made it public.
The unusual detail is not the size. It is the chip. Almost every other major AI infrastructure deal of the past two years is denominated in GPU capacity: Nvidia H100s, H200s, or the Blackwell generation. Amazon built its Trainium processor to handle Anthropic’s Claude training workloads. Google developed custom TPUs for the same purpose. The Akamai contract is structured entirely around central processing unit workloads. CPUs, not graphics processors.
Anthropic has been assembling a compute portfolio that no single hyperscaler can supply. Earlier this year it signed a $10 billion data center deal with Volta Infra in Norway for hydroelectric AI infrastructure. Its commitment to Amazon Web Services runs into the tens of billions. What Akamai offers is different in kind: a network of more than 4,000 distributed edge nodes across 135 countries, built over two decades to move content close to users. For AI inference workloads that require low latency, an agent responding to a request in real time rather than a model training overnight, distributed edge compute offers something a centralized data center cannot replicate.
The equity terms are the part of the contract that rewrites the vendor relationship entirely. Akamai issued Anthropic a warrant to purchase non-voting convertible Series B Preferred Stock representing up to 7.7 million shares on an as-converted basis, approximately 5 percent of Akamai’s outstanding common stock, at an exercise price of $111.33 per share, as Akamai disclosed in its announcement Wednesday. The vesting schedule is tied directly to contract performance: 2 percent vests on the $11.6 billion commitment, the remaining 3 percent vests in roughly 1 percent increments for each additional $3 billion Anthropic commits beyond that. The infrastructure provider is paying its customer in equity to keep using its infrastructure.
Dario Amodei, Anthropic’s chief executive, has argued publicly since September that frontier AI labs need more external oversight of their systems. The warrant structure creates a different kind of alignment: Anthropic’s financial incentives are now tied to Akamai’s stock performance, and Akamai’s stock performance is now tied to Anthropic’s infrastructure spending decisions. The deal is not a governance mechanism. It is the opposite: it deepens the financial interdependence between a frontier lab and its infrastructure provider.
The math describes the actual risk. Akamai is spending almost two billion dollars this year on infrastructure before Anthropic pays for it. The company’s stock may have jumped 20 percent on the announcement, but the financial clock on the contract does not start for more than a year. Tom Leighton, Akamai’s chief executive, has described the company’s strategy as a deliberate transformation from content delivery network into AI infrastructure provider. Wednesday’s announcement is the largest single test of whether that transformation generates the revenue it requires.
Microsoft, Amazon, Google, and Oracle have been absorbing AI compute demand at a scale that requires tens of gigawatts of new data center capacity, more than they can supply alone. CoreWeave built a multi-billion-dollar business filling GPU gaps the hyperscalers could not meet. Akamai appears to be making the equivalent bet for CPU-based inference workloads: the AI compute demand that requires distributed global reach rather than the raw parallel processing power of a GPU cluster. Whether that market is as deep as Anthropic’s $11.6 billion commitment suggests will be clearer by 2027.
The contracts governing Accenture’s role as Anthropic’s embedded safety evaluator specify reporting obligations that cover model deployment infrastructure. Whether Akamai’s distributed CPU network falls within that oversight scope, whether a third-party infrastructure provider sits inside or outside the safety perimeter Anthropic has publicly committed to, was not addressed in Wednesday’s announcement. The financial exposure is clearly disclosed. The safety governance structure for what could become a $20 billion AI infrastructure relationship is not yet.
