I’ve lost count of the KubeCon interviews where the talk turns to “AI is hard”, but few cut through the noise like my quick chat with Gary Gaughan, Akamai’s product manager for LKE and LKE Enterprise, right there on the Atlanta show floor.
Amid the booth excitement and echoes of the recent keynote, Gary laid out how Akamai’s managed Kubernetes is evolving to handle the real pains of edge AI: scarce GPUs, observability black holes, and the scramble for low-latency inference that doesn’t route half your traffic back to a central region.
If you’re knee-deep in scaling ML workloads across clusters that feel more like a federation of fiefdoms than a unified platform, this quick chat is your coffee break read (or watch). It’s less about flashy demos and more about the community-driven approach that keep production humming.
The Community Swarm: Solving AI’s Messy Problems
Gary kicked off with what makes KubeCon tick: the open-source goodness. We chatted about how new projects bubble up to tackle AI’s curveballs, like sandboxing models that could leak data across tenants or tracing spans that vanish when your pods hop PoPs. For Akamai, that means baking Kubernetes standards into their upcoming releases, ensuring LKE (their managed K8s) plays nice with the CNCF ecosystem without forcing you to reinvent observability wheels.
The big unlock? GPUs. “You must have access to those critical GPUs,” Gary stressed, highlighting Akamai’s ramp-up on GPU diversity and broad access in their data centers. We’re not talking about your typical consumer gear. These are production-grade resources for training and inference, co-located with Akamai’s network to slash the latency tax on edge use cases. Think fraud detection that doesn’t wait for a round-trip to Virginia, or recommendation engines that feel instant whether you’re in Paris or Poughkeepsie.
App Platform: Edge AI Without the Egress Hangover
A standout from last year’s KubeCon, and now battle-tested in production, is Akamai’s App Platform. It’s a one-stop for spinning up projects, from basic services to full AI inference pipelines, with “really short time to value,” as Gary put it. Deploy a Hugging Face model? Done in minutes, pushed to the edge for sub-100 ms responses.
What sets it apart: Akamai’s 4,400 global Points of Presence (PoPs). “The joining of the Akamai network… with really low latency delivery is really allowing our customers to execute new use cases that are maybe not possible on other hyperscalers,” Gary explained in conversation at the event. No more hyperscaler lock-in where your “global” app funnels through three regions. Here, it’s distributed from the jump which is perfect for solving data sovereignty headaches or IoT telemetry that needs to stay local.
Here’s a quick snapshot of the edge cases Gary flagged as prominent capabilities that customers are benefitting from:
| Use Case | Why It Wins on Akamai | Ops Perk |
|---|---|---|
| AI Inference at Edge | Low-latency model serving via PoPs | No central routing; <100 ms P99 typical |
| GPU Workload Mgmt | Blackwell access in Akamai DCs | Scarce resource allocation without queues |
| Observability Swarm | CNCF standards for multi-cluster traces | Fewer blind spots across 4,400+ PoPs |
These aren’t pie-in-the-sky numbers; they’re from customer stories and proven in the field where teams ditched custom hacks for Akamai’s pillars: delivery, continuous delivery, and security.
From the Ops View
I’ve stared down enough Grafana alerts at midnight to know AI isn’t just “add more pods.” It’s about primitives that scale without exploding your bill…or your sanity. Gary’s take reinforces what I saw at the Fermyon booth last month: Akamai’s stacking the deck for edge-native stacks that treat Kubernetes as the control plane, not the bottleneck.
“We can’t do what we do without all the community effort,” he wrapped up, and that’s the signal for 2026. Positive feedback from customers means demand’s spiking, and if you’re architecting for distributed AI, Akamai’s LKE feels less like a side bet and more like the safe harbor.
You can also watch the interview here. it’s a reminder that the best tech talks happen in the hallways, not just the keynotes. Then spin up an LKE cluster and test that GPU access yourself. Your SLOs will thank you.
Don’t forget to also check out our coverage of the Fermyon acquisition and more from Akamai at KubeCon 2025.
