Anthropic commits $11.6 billion to Akamai for CPU cloud compute

Overview

Updated 43 minutes ago

Anthropic committed to spend $11.6 billion over seven years on Akamai's cloud, a contract announced Sept. 24. The striking detail: the money buys CPU capacity, not GPUs.

For years, serious AI infrastructure meant dense GPU clusters. This deal is the clearest sign yet that inference workloads are splitting from training, and that Akamai's 4,400-site distributed network is where agent-heavy CPU work lands. Akamai plans roughly $5.5 billion in capital spending to build the capacity, with revenue not expected until mid-2027.

Why it matters

If this deal holds, a slice of AI's future runs on distributed CPUs, not just pricey GPU clusters — changing who profits from AI infrastructure.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

$11.6B

Committed contract value

Anthropic's seven-year commitment to Akamai, the largest deal in Akamai's history.

$20B

Potential total value

Up to $9 billion more in mutually agreed tranches of $3 billion each.

$5.5B

Planned Akamai capital spending

$1.7 billion in late 2026 for components, $3.1 billion in 2027, $700 million in 2028.

Akamai warrant stake for Anthropic

Up to 7.7 million shares at $111.33; about 2% vests now, 1% per additional $3 billion spent.

$1.7B

Expected annual revenue run-rate

Akamai expects to reach this pace by end of 2028, with $150-300 million in 2027.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

People Involved

Ed McGowan

Chief financial officer, Akamai

Led the investor call explaining the deal's economics

Organizations Involved

Akamai Technologies

Cloud and edge infrastructure company

Selling $11.6 billion of distributed cloud capacity to Anthropic

A content-delivery network operator repurposing its 4,400-location distributed network for AI compute.

Anthropic

Artificial intelligence company

Committed $11.6 billion for CPU compute; holds warrant for up to 5% of Akamai

The AI lab behind the Claude model family, and one of the largest buyers of cloud compute.

Timeline

May 2026 September 2026

4 events Latest: 4 days ago

Tap a bar to jump to that date

September 2026
Deal announced publicly
Latest Statement

Akamai reveals the $11.6 billion commitment, expandable to $20 billion, and grants Anthropic a warrant for up to 5% of its stock.
4 days ago
Project Plans 2 and 3 signed
Agreement

Anthropic commits to pay the full $11.6 billion over seven-year terms, subject to delivery and availability requirements.
September 18th, 2026
May 2026
Master services agreement signed
Agreement

Akamai and Anthropic sign the master agreement that later hosts the project plans.
May 5th, 2026
Bloomberg reports $1.8 billion deal
Report

Media reports describe an earlier, smaller cloud deal between the two companies.
May 1st, 2026

Scenarios

Deal expands to $20 billion as Anthropic exercises the option

Possible Resolves by End of 2028

Discussed by: Akamai executives and analysts covering AI infrastructure

If agent-driven CPU demand keeps climbing, Anthropic buys the additional three $3 billion tranches. Each purchase vests another 1% of Akamai's stock warrant. Akamai's revenue run-rate reaches $1.7 billion annually by end of 2028, and the full commitment plays out through the early 2030s. This outcome depends on CPU capacity staying cost-competitive with GPU alternatives.

Expansion stalls; deal holds at $11.6 billion

Possible Resolves by End of 2028

Discussed by: Financial analysts noting the option depends on mutually agreed terms

Anthropic uses the committed capacity but never adds the extra $9 billion. The warrant stays at the roughly 2% that vested at signing. Akamai fills spare CPU capacity with other customers. Revenue follows the announced path — $150-300 million in 2027, ramping to about $1.7 billion annually — but no second wave comes.

Deliveries falter; Anthropic trims or exits

Unlikely Resolves by End of 2027

Discussed by: Reports noting the contract's service-availability and termination clauses

The agreement is not ironclad. Either side can end it under certain conditions, and Anthropic holds termination rights tied to uncured breaches and material outages. If Akamai misses delivery or availability targets, Anthropic could reduce or cancel a project plan. The $1.7 billion already committed to memory components would then sit against weaker demand.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

January 2023

Microsoft-OpenAI Azure commitment (2023)

Microsoft invested $13 billion in OpenAI and agreed to provide massive Azure capacity to support the lab's models, effectively making Microsoft the primary compute vendor for a frontier AI company.

Then

OpenAI gained the compute it needed to scale ChatGPT and later models, while Microsoft locked in a stake in a leading lab.

Now

The relationship proved fragile; by 2025 reports described the capacity deal being renegotiated as OpenAI's needs outgrew the original terms.

Why this matters now

Big capacity commitments between AI labs and infrastructure vendors get renegotiated as demand shifts. The Akamai-Anthropic deal carries the same expansion, revision, or exit risk.

2025

Oracle's AI capacity deals (2025)

Oracle signed enormous multiyear capacity agreements with OpenAI and other AI companies to supply cloud compute for training and inference, becoming a major alternative to the largest hyperscalers.

Then

Oracle's cloud backlog ballooned as AI labs committed to hundreds of billions in future capacity.

Now

The scale of these commitments raised questions about whether demand would materialize as promised and how quickly labs would actually consume the capacity.

Why this matters now

It shows how AI labs are spreading compute commitments across multiple vendors. The Akamai deal extends that pattern down-market, to distributed CPU capacity rather than centralized GPU clusters.