Anthropic has announced a landmark $11.6 billion, seven-year agreement with Akamai Technologies for CPU computing capacity, the largest CPU infrastructure deal in cloud computing history. The contract can expand by an additional $9 billion, potentially reaching $20 billion over the agreement’s lifetime.
The deal includes an equity warrant for Anthropic to acquire up to 5% of Akamai at $111.33 per share, aligning the two companies’ incentives. Akamai expects the capital expenditures related to this commitment to total approximately $5.5 billion.
Why CPUs, Not GPUs?
The deal’s focus on CPU capacity rather than GPUs is notable. While most AI infrastructure spending has historically prioritized GPU clusters for model training, Anthropic’s CPU commitment signals a strategic pivot toward inference and agent workloads—the compute-intensive tasks that happen when AI models actually run in production.
“Agents make many more API calls than static chat interactions,” noted one infrastructure analyst. “A single agentic workflow might involve hundreds of tool calls, each requiring compute. CPUs are cost-effective for this pattern.”
This represents a broader industry trend: as AI shifts from passive chatbots to autonomous agents that take action, the infrastructure demands are fundamentally different. Agentic AI requires rapid, stateless computation across diverse tool integrations—a workload where CPU efficiency often outperforms GPU acceleration on a per-dollar basis.
Industry Implications
The Anthropic-Akamai partnership marks a turning point in how AI companies approach infrastructure. Rather than building proprietary data centers, AI labs are increasingly partnering with established cloud and edge providers for specialized workloads.
Akamai, traditionally known for content delivery and cybersecurity, has positioned itself as a “compute-first” cloud alternative, particularly for inference and agentic workloads. The Anthropic deal validates this strategy and could spark similar agreements across the industry.
For enterprises deploying Claude and other AI assistants at scale, the deal suggests that inference costs may drop significantly as specialized CPU infrastructure scales—potentially accelerating the shift from pilot programs to production agent deployments.