Two quarters into the four I gave Akamai in March to show whether its artificial intelligence infrastructure push was a business or a bet, one customer has answered with $11.6 billion over seven years (Akamai Technologies, "Akamai Announces"). Akamai announced the agreement with Anthropic on September 24, and its shares rose more than 20% in after-hours trading (Riley).

The confirmed Anthropic sequence starts in May, not March. A $200 million, four-year enterprise agreement in March involved a separate, unnamed customer and belongs to Akamai's own AI Grid launch, not to Anthropic (Akamai Technologies, "Akamai Launches AI Grid"). Bloomberg reported that Anthropic sat behind a $1.8 billion commitment Akamai disclosed in May (Riley). The September figure adds to more than $2.8 billion of multi-year cloud commitments Akamai had announced earlier in the year, across its full customer base.

Akamai's original insight applies to AI traffic too

Akamai built its network on one observation in the late 1990s: content sitting far from users creates delay, and moving copies closer removes it. The same logic governs AI connectivity between endpoints and data centers, and Akamai already operates the distributed network that reaches both. Anthropic locking in $11.6 billion and picking up equity as it spends follows from that: protect the pipeline you will depend on for the next seven years.

This contract buys CPU capacity while Akamai's graphics processing unit pipeline is sold out

Akamai describes the workloads as central processing unit demand. Akamai management said in August that its GPU pipeline was sold out, and it paused share repurchases to fund that build ("Akamai (AKAM) Q2").

Agentic products run generated code and wait on tool calls between model calls, and that work lands on CPUs. That reading is my inference from how agent systems are built, and neither company has confirmed it. If agents are the workload, GPU allocation is no longer the only capacity question a buyer of AI services has to ask.

If agents are the workload, GPU allocation is no longer the only capacity question a buyer has to ask.

Akamai's services line up with what Anthropic's enterprise customers need

Akamai sells three things: cloud computing, security, and content delivery. Inside the cloud tier, Akamai Inference Cloud runs AI workloads across its distributed footprint, and CPU and GPU compute sit alongside Kubernetes and object storage as line items enterprises can provision on their own. Inside security, App & API Protector, Firewall for AI, Bot Manager, and Guardicore Segmentation exist to keep an enterprise's exposed endpoints from becoming an attack surface. Inside content delivery, the edge network that once cached web pages now routes traffic across more than 4,400 locations.

Line those up against what an enterprise running Claude in production needs, and the fit is not incidental. Agent traffic between a company's own systems and Anthropic's models benefits from a network built to sit close to both ends. An enterprise's API endpoints, the ones an agent calls to take action, need the same bot and firewall protection Akamai already sells. Neither company has said the Anthropic contract routes through these specific products, and that connection is mine, not a claim either company has made.

Akamai built capability on its own before this contract arrived

In March, Akamai launched Inference Cloud as the first global-scale deployment of the NVIDIA AI Grid reference design, routing AI workloads across more than 4,400 edge locations (Akamai Technologies, "Akamai Launches AI Grid"). In May, Akamai agreed to acquire LayerX, extending its Zero Trust portfolio into the browser so security teams can see how employees use generative AI tools without switching browsers (Akamai Technologies, "Akamai Technologies Announces Intent to Acquire LayerX"). Both are Akamai's own strategy, not contracts with Anthropic.

The confirmed relationship with Anthropic runs on a separate, shorter track. Bloomberg's May report ties Anthropic to the $1.8 billion commitment. That same month, Akamai joined Anthropic's Project Glasswing, testing Claude Mythos Preview against its own codebase to find vulnerabilities before attackers do (Maudsley). That makes three confirmed items between the two companies since May: the cloud commitment, the Glasswing collaboration, and this contract. LayerX and AI Grid are Akamai's own moves, not part of that count.

A cloud contract at Akamai moves memory purchasing a layer down

Anthropic has lined up capacity from several suppliers this year, including a 20-year lease for about 401 megawatts at a TeraWulf campus in Hawesville, Kentucky, and a June supply agreement with Micron Technology for memory and storage (Riley). Anthropic's demand now reaches the memory market twice, through its own Micron agreement and through the pre-purchase Akamai ties to this contract.

A chief information officer who reads the announcement as a cloud contract sees one layer. Akamai's memory pre-purchase moves a second layer, which is the dependency chain behind why I cover the entire AI stack.