Akamai Launches AI Inference and Agent Sandbox to Expand Cloud AI Services

What Akamai Is Actually Offering

Akamai has been a major player in edge networking and CDN for years. The move into AI model operations is a natural extension of that infrastructure — but it’s a meaningful step beyond just selling GPU capacity.

These two products add a service layer on top of Akamai’s existing cloud and edge resources, targeting enterprises that want to deploy AI in production without managing the underlying complexity themselves.

Akamai Inference: Deploy Models Without the Server Headache

Akamai Inference is a serverless AI model deployment service. Enterprises can run publicly available open-weight models or their own fine-tuned versions directly in production applications.

Two deployment modes are available:

• Serverless — pay based on usage, no dedicated resources required
• Dedicated — allocate specific resources for consistent, predictable performance

The service currently supports 12 open-weight models, including OpenAI’s gpt-oss-120b, Google’s Gemma 4, and Nvidia’s Nemotron 3 Ultra. It also supports embedding models for retrieval-augmented generation (RAG), which is increasingly important for enterprises that need AI to reference internal documents and proprietary data.

One practical detail worth noting: Akamai Inference uses an OpenAI-compatible API. That means teams can switch between supported models without rewriting their existing application code — a real friction reducer for teams already building on OpenAI-style interfaces.

Akamai Agent Sandbox: Isolated Execution for Agent-Generated Code

AI agents that write and execute code introduce a specific security problem: you need that code to run, but you don’t want it touching systems it shouldn’t.

Akamai Agent Sandbox addresses this with WebAssembly-based isolation. Code generated by AI agents runs in a contained environment with restricted access to compute resources, network connections, and system functions. Akamai also plans to add more granular agent permission controls going forward.

This is a practical answer to a real concern that’s slowing enterprise adoption of agentic AI workflows. Isolated execution doesn’t eliminate risk, but it significantly reduces the blast radius if something goes wrong.

Why Akamai’s Edge Network Matters Here

Akamai operates more than 4,350 edge locations globally. For AI inference specifically, that footprint means requests can be processed closer to end users — which translates to lower latency in production environments.

This is a meaningful differentiator compared to centralized cloud AI services where all inference traffic routes through a handful of data centers. For latency-sensitive applications, edge-based inference is worth paying attention to.

Who This Is Actually For

These services are positioned for enterprise teams that:

• Want to run open-weight models in production without managing GPU infrastructure
• Are building RAG pipelines and need embedding model support alongside inference
• Are deploying AI agents and need secure, isolated code execution
• Already use OpenAI-compatible APIs and want flexibility to switch models

It’s less relevant for teams running lightweight experiments or startups that are still in early prototyping. The value proposition sharpens at production scale.

The Bigger Picture

Akamai’s COO Adam Karon framed the goal clearly: customers want to build, run, and protect AI on a single platform. These two products are designed to add inference and execution layers to what Akamai already offers in cloud storage, networking, and compute.

The “single platform” angle is worth watching. Enterprises are increasingly fatigued by stitching together multiple vendors for AI infrastructure. If Akamai can credibly cover model deployment, secure agent execution, and edge delivery under one roof, that’s a consolidation story that will resonate with enterprise buyers.

Both services are in beta, so expect the feature set — particularly around agent permission controls — to evolve. If you’re evaluating AI infrastructure options for production workloads, Akamai Inference and Agent Sandbox are worth putting on your comparison list now rather than waiting for general availability.