Akamai pushes edge + GPUs

Akamai is trying to turn its edge network into an artificial intelligence platform by pairing managed Kubernetes with graphics processing units and nearby compute. (tfir.io) The basic idea is simple: run the first, delay-sensitive parts of an artificial intelligence job near users, then send heavier work to larger clusters. Akamai’s Linode Kubernetes Engine is the software layer that schedules those containerized workloads across that spread of infrastructure. (techdocs.akamai.com) (tfir.io) Graphics processing units, or chips built to do many calculations at once, are the hardware behind that pitch. Akamai says its cloud now offers Nvidia RTX Pro 6000 Blackwell Server Edition, Nvidia RTX 4000 Ada, and Quadro RTX 6000 options, with plans that can scale to eight cards per instance. (techdocs.akamai.com) Akamai started pushing this architecture into products in 2025. On March 27, 2025, it launched Akamai Cloud Inference and said customers could see 3x higher throughput, up to 2.5x lower latency, and as much as 86% lower cost than traditional hyperscaler setups. (prnewswire.com) The company had already laid part of the groundwork a month earlier. On February 25, 2025, Akamai announced a managed container service that it said spanned more than 700 cities and over 4,300 points of presence on its cloud platform. (prnewswire.com) That matters because many artificial intelligence applications do not fail on model quality first; they fail on delay. Akamai executives told The New Stack this month that robotics, fraud detection, and conversational agents are the kinds of workloads where slower round trips can quickly hurt the user experience. (thenewstack.io) Akamai is also drawing a line between central and distributed infrastructure instead of claiming the edge replaces big data centers. In its 2025 launch, the company said large language model training would stay in hyperscale facilities, while inference and other “actionable work” would move closer to users. (prnewswire.com) The network footprint is the selling point. Akamai’s executives said this month that the company runs 41 core data centers in 36 countries and extends that with about 4,400 smaller distributed-reach data centers worldwide. (thenewstack.io) The company has been adding more enterprise controls around that footprint. Akamai says Linode Kubernetes Engine Enterprise supports up to 500 nodes and 5,000 pods, with a highly available dedicated control plane and Cilium networking for higher-performance cluster traffic. (techdocs.akamai.com) Nvidia is central to the hardware stack as well as the message. When Akamai expanded its inference cloud on October 29, 2025, it said the service used Nvidia RTX Pro servers, BlueField-3 data processing units, and Nvidia AI Enterprise software, with initial availability in 20 locations. (datacenterdynamics.com) Investors have started to see some of that cloud push in the numbers, even if it is still small beside Akamai’s larger businesses. In second-quarter 2025 results released on August 7, 2025, Akamai said Cloud Infrastructure Services revenue reached $71 million, up 30% from a year earlier. (ir.akamai.com) The bet is that companies will buy one platform for both proximity and orchestration, rather than stitch together separate edge, cloud, and artificial intelligence stacks. Akamai has spent the past year lining up the pieces — containers, graphics processing units, and edge locations — so that argument is now a product pitch instead of a roadmap. (tfir.io)