What did Anthropic and Akamai agree to?

On September 18, 2026, Anthropic and Akamai Technologies entered into two new project plans under an existing Master Services Agreement. The combined commitment is approximately $11.6 billion over seven years, with an option to expand by up to $9 billion more, bringing the total potential value to $20 billion. Akamai will supply CPU-based cloud infrastructure and distributed software to support Anthropic's growing workloads.

The Akamai deal is specifically for CPU workloads, not GPU clusters. Frontier AI labs need massive CPU capacity alongside GPU training farms: CPUs handle inference orchestration, routing, tokenization, safety filtering, memory lookups, and agentic control loops. Akamai's globally distributed edge network gives Anthropic low-latency serving at scale for tasks that do not require accelerator silicon.

What is the warrant Anthropic received from Akamai?

Akamai issued Anthropic a warrant to purchase up to 387,051 shares of Series B non-voting convertible preferred stock, equivalent to approximately 7.7 million shares of common stock, or about 5% of Akamai's outstanding shares. The exercise price is $111.33 per common share equivalent. Approximately 2% vests with the $11.6 billion commitment; the remaining 3% vests as Anthropic commits each additional $3 billion of expansion, up to the full $9 billion.

When does Akamai start earning revenue from the deal?

Akamai expects no revenue from the Anthropic contract in 2026. Revenue begins in the second half of 2027 in the range of $150 million to $300 million for the full year. By the end of 2028 the relationship is expected to be fully ramped, reaching an annualized run rate of approximately $1.7 billion per year.

What does this mean for enterprise AI buyers?

Anthropic's compute buildout signals long-term cost investment to support Claude at scale. For enterprise buyers, it suggests Anthropic is building the infrastructure to sustain low-latency, high-volume agentic workloads without depending solely on AWS or Google Cloud. It also shows that frontier AI pricing can accommodate large infrastructure costs and still trend downward, as Anthropic's compute efficiency improves.

Anthropic's $11.6B Akamai Deal and the CPU Compute Shift

Anthropic signed the largest infrastructure contract in Akamai’s history on September 24, 2026: an $11.6 billion, seven-year agreement for CPU-based cloud capacity, with terms that can expand to $20 billion. Akamai’s stock jumped 15 to 22 percent in after-hours trading on the news.

For enterprise AI buyers, the deal is not simply a procurement announcement. It maps Anthropic’s infrastructure strategy in the months before its planned IPO, and it raises a pointed question about how frontier labs are actually wiring together the compute they need to run models at enterprise scale.

What Akamai Is Providing (and Why CPU Matters)

The contract is explicitly for CPU workloads, not GPU training clusters. That distinction matters.

Most coverage of AI compute fixates on Nvidia H100s, H200s, and GB200s because those chips do the heavy lifting in model training. But running a model in production, especially a large agentic model that needs to handle thousands of parallel sessions, involves an enormous amount of CPU work: tokenization, safety classification, routing across model variants, memory retrieval, agent state management, and orchestration between sub-tasks.

Akamai brings something specific to this: a globally distributed edge network spanning thousands of points of presence. Rather than serving all Claude requests from a handful of hyperscaler data centers, Anthropic can push latency-sensitive CPU work closer to the end user. For real-time agentic applications, where response latency compounds across multi-step workflows, that geography matters.

Akamai CEO Tom Leighton described the deal as “the largest contract in our company’s history” in a call with analysts. Akamai expects to spend approximately $5.5 billion in capital to fulfill the commitment, with $1.7 billion going out the door in 2026 to pre-purchase memory and other critical supply chain components.

Anthropic’s Compute Accumulation Before IPO

The Akamai deal is the latest in a string of major compute commitments Anthropic has made in 2026. Taken together, they paint a picture of a company building infrastructure independence ahead of its anticipated stock market debut.

Partner	Reported Value	Type	Announced
Akamai	$11.6B (up to $20B)	CPU cloud infrastructure	September 2026
Rum Group	$13.7B	Data center capacity	September 2026
Volta	$10B	AI cloud computing	August 2026
Nscale	$45B	West Virginia campus	August 2026
Akamai (initial)	$1.8B	Cloud infrastructure	May 2026

These deals span a range of infrastructure types: dedicated GPU clusters for training (Volta, Nscale), large data center campuses (Rum Group), and now distributed CPU capacity for serving and orchestration (Akamai). The pattern is diversification. Anthropic is not concentrating its infrastructure on a single hyperscaler. It has existing agreements with AWS and Google Cloud, and it is adding a parallel layer of owned and contracted capacity that it controls more directly.

For a company preparing to go public at a reported $2 trillion valuation, this makes strategic sense. Gross margins in AI software depend partly on keeping inference costs below a threshold where they eat into model revenue. A company that can negotiate multi-year bulk infrastructure deals, capture equity warrants in its suppliers, and distribute workloads efficiently has better cost leverage than one renting capacity month to month.

The warrant structure in the Akamai deal is particularly notable. Anthropic receives the right to buy approximately 5% of Akamai’s common stock at $111.33 per share. About 2% vests immediately with the $11.6 billion commitment; the rest vests as Anthropic expands the relationship toward $20 billion. This aligns the financial interests of both parties and gives Anthropic potential upside if Akamai’s cloud business grows as a result of AI demand.

The CPU Demand Signal

There is a broader signal here for enterprise AI builders.

As agentic AI deployments grow in complexity, the compute architecture shifts. Early AI deployments were simple: one model call, one response. Modern enterprise deployments involve orchestration layers, tool calls, memory retrieval, safety checks, multiple model variants for different cost and quality tiers, and parallel background agents. Most of that surrounding work runs on CPU, not GPU.

Anthropic’s decision to sign an $11.6 billion CPU deal suggests the company sees this demand growing substantially. From an enterprise architecture perspective, it validates the idea that AI infrastructure planning should include both GPU capacity (for fine-tuning and heavy inference) and CPU capacity (for orchestration, routing, and real-time agentic workloads).

For teams building on Claude today, the Akamai infrastructure will not be directly visible. But it will shape the latency, reliability, and geographic availability of Claude-based applications as Akamai’s deployment ramps through 2027 and 2028. Revenue from the deal begins in the second half of 2027, giving a rough timeline for when the new infrastructure layer starts to influence Claude’s serving characteristics.

What This Means for Enterprise AI Strategy

Several implications follow from this deal.

Infrastructure independence is a moat. Anthropic is not waiting for AWS or Google to expand capacity on its behalf. It is signing direct deals at a scale that gives it leverage over supply chains, pricing, and deployment timelines. Enterprise AI teams that are similarly dependent on a single cloud provider for AI infrastructure should take note: concentration risk is a real concern as model demand grows.

CPU is not a footnote. The assumption that AI is a GPU story misses a large fraction of the workload. Enterprises deploying agentic systems should plan their CPU and memory budgets with the same rigor they apply to GPU allocation.

Warrant structures are a new pattern. Anthropic received equity in Akamai as part of the compute deal. This is not a standard vendor relationship. It reflects a world where frontier AI labs have enough purchasing power to negotiate strategic alignment with their suppliers. For enterprise AI teams building vendor relationships, understanding who has these equity stakes can clarify whose long-term interests align with yours.

Cost trends remain favorable. Despite committing tens of billions in infrastructure spending, Anthropic has continued to reduce API prices. Claude Opus 5.5, released September 22, costs $4 per million input tokens and $20 per million output tokens. The compute investments being made now are intended to drive those numbers lower over time, not higher. Enterprises locking in contract structures today, before the IPO repricing that often follows public listings, have an opportunity to secure favorable unit economics.

The Anthropic-Akamai deal is the largest of its kind for Akamai and one of the largest infrastructure commitments in AI’s short history. It is also a concrete sign that the compute race underlying frontier AI development is far from over. For teams building AI-native operations, understanding where that compute is going, and why, is part of understanding what you can rely on, and when.

For a deeper look at how enterprise teams are structuring their AI operations around models like Claude, see our guide to autonomous GTM systems.

If you want to talk through how this affects your enterprise AI stack, book a call with the Enera team.