Akamai pushes AI to the edge
Akamai has launched an 'AI Grid' to orchestrate distributed inference across some 4,400 edge sites using NVIDIA GPUs, aiming to reduce latency and operational cost for AI workloads. The move highlights a shift in how AI services can be deployed outside centralized cloud datacenters. (edgeir.com)
Artificial intelligence usually runs in big, centralized data centers. Akamai is now spreading that work across 4,400 edge sites so responses can be generated closer to users and devices. (edgeir.com) Akamai said on March 16 that its Akamai Inference Cloud is the first global-scale implementation of NVIDIA AI Grid, a system for routing inference jobs across edge, regional, and core infrastructure. The company said the network is being equipped with thousands of NVIDIA RTX PRO 6000 Blackwell Server Edition graphics processors. (globenewswire.com) Inference is the stage when a trained model answers a prompt, labels a video frame, or helps software make a decision in real time. Akamai said pushing that step to the edge cuts the “round trip to a centralized cluster” for uses such as real-time video, gaming, and personalized services. (edgeir.com) Akamai first introduced Inference Cloud on October 28, 2025, saying it would extend artificial intelligence from core data centers to the edge of the internet. At that launch, the company said its global edge network had more than 4,200 locations and combined NVIDIA servers, BlueField-3 data processing units, and NVIDIA AI Enterprise software. (prnewswire.com) The company’s pitch is that not every artificial intelligence job belongs in an “AI factory,” the industry term for giant centralized clusters built for training frontier models. Akamai said centralized systems still make sense for training, while edge inference is better suited to applications that need predictable latency at the point of contact. (datacenterdynamics.com) NVIDIA is making the same argument more broadly with telecom and network operators. In a March 17 post tied to NVIDIA GTC 2026, the company said operators including AT&T, T-Mobile, Comcast, Spectrum, and Akamai are turning distributed network sites into “AI grids” that run inference closer to users, devices, and data. (nvidia.com) Akamai is also trying to show that customers will pay for this mix of centralized and distributed compute. On March 5, the company disclosed a four-year, $200 million service agreement with a major United States tech company that will use a multi-thousand NVIDIA Blackwell graphics processor cluster hosted in an Akamai data center. (ir.akamai.com) That deal points to the model Akamai is building: large clusters in core facilities for heavier workloads, with edge sites handling time-sensitive inference closer to the user. If enterprises buy into that split, Akamai’s old content-delivery footprint could become part of the infrastructure layer for artificial intelligence services. (ir.akamai.com)
