Executive Summary
Anthropic has signed an $11.6 billion compute capacity deal to lease from Akamai Technologies. The agreement focuses on CPU workloads rather than GPU capacity, indicating a strategic shift toward inference and auxiliary processing infrastructure. This deal underscores the maturation of AI infrastructure economics beyond training-focused clusters.
The Players
Anthropic, the AI safety-focused company behind Claude, is the buyer. Akamai Technologies, traditionally known for content delivery network (CDN) services, is the provider. Akamai has been repositioning itself as a cloud and edge computing infrastructure company, and this deal validates that strategic pivot at scale.
The Numbers
$11.6 billion in committed compute capacity. The lease is set for seven years and includes expansion up to $9 billion, for a possible total commitment of $20 billion.
So What
This deal matters for three reasons:
First, it confirms that inference economics are driving massive infrastructure commitments. Training gets the headlines, but inference is where sustained, predictable load growth lives. CPU-based inference workloads create different power demand profiles: lower peak density than GPU clusters, but higher utilization rates and longer-duration baseload demand.
Second, it validates alternative infrastructure providers. Akamai isn’t AWS, Azure, or GCP. This deal proves that AI companies are willing to commit billions to non-hyperscaler infrastructure if the economics and capacity are there.
Third, for power developers and investors: CPU inference loads are ideal PPA counterparties. They need reliable, cost-effective baseload power with high uptime requirements. This is exactly the demand profile that makes dispatchable generation and long-term power purchase agreements pencil.
What to Do
If you’re developing generation assets, start modeling CPU inference loads as distinct from GPU training loads. Different power density, different reliability requirements, different contract structures. If you’re financing data center infrastructure, watch for more deals like this — the inference build-out is just beginning, and it needs different infrastructure than training. And if you’re Akamai’s competitors, this just set the benchmark for what AI companies are willing to commit for the right capacity.
Source: Data Center Dynamics, September 28, 2026
