(SeaPRwire) – By: Ethan Gallagher
Anthropic just signed a seven-year, $11.6 billion infrastructure contract with Akamai. This is not a standard cloud deal. It is a structural bet that the center of gravity in AI compute is shifting from training to inference. The headline numbers matter, but the underlying architecture choice matters more. For years, the industry obsessed over GPU scarcity for model training. Now, the largest AI labs are looking elsewhere to run the code.
The press release states the facts clearly. Anthropic committed to buying CPU-based cloud infrastructure from Akamai. The base value is $11.6 billion. It can expand to $20 billion. Anthropic also received a warrant for up to 5% of Akamai’s stock. The purchase price is set at $111.33 per share. The deal generates about $1.66 billion in annual recurring revenue for Akamai. This is not a one-off experiment. It is a multi-year revenue lock-in for the infrastructure layer.
The subtext here is critical. Akamai is providing CPU workload support, not H100 or A100 clusters. Training large language models requires massive GPU parallelism. Running those models in production, known as inference, is different. Inference is latency-sensitive and often relies on general-purpose compute for pre-processing, routing, and lightweight tasks. By locking in CPU capacity, Anthropic is optimizing for the operational cost of serving AI to users, not just building the next model version. This separates the “builder” spend from the “runner” spend.
The industry narrative has been dominated by NVIDIA’s stock surge. Everyone assumes AI growth equals GPU growth. This deal suggests a divergence. The physical scaling limits of cooling and power delivery are hitting a wall. CPU-dense racks are cheaper to maintain and deploy at the edge. Akamai’s global footprint allows for distributed inference. This reduces latency for end users. The hardware supply chain is about to fragment. You will have specialized silicon for training and commodity hardware for serving. The margins will shift accordingly.
Don’t expect the GPU narrative to die immediately. But the next wave of infrastructure spending will look different. Watch the capital expenditure lines in Q4 reports. If you see a rise in CPU-centric data center investments alongside GPU clusters, the shift is confirmed. The market has already reacted to this news, with analysts raising price targets to $185 and $225. The stock price is now a reflection of this new paradigm. The era of pure GPU dependence is ending. The era of hybrid compute is here.
Author bio: Ethan Gallagher, a Silicon Valley Hardware Architect and Infrastructure Strategist, specializes in analyzing the physical and economic constraints of next-generation data center architectures.