Akamai has launched two managed services for its AI Cloud: Akamai Inference, which serves open-weight and custom models through OpenAI-compatible APIs, and Akamai Agent Sandbox, an isolated environment where AI agents can run code inside enforced boundaries. Why it matters: Akamai is turning its global edge network — more than 4,350 locations — into a full AI stack for enterprises that want to run models and agents without assembling the pieces themselves.
October 27, 2026
What happened
Akamai announced the two products on Tuesday, October 27. Akamai Inference gives enterprises managed access to models without operating the serving stack themselves: customers can choose serverless, pay-per-use inference or dedicated deployments, and existing applications can move over without rewrites because the APIs are OpenAI-compatible. Both products are available in beta, and the Inference beta includes the serverless options.
What Akamai Inference includes
At launch, the catalog includes 12 open-weight models ranging from 8 billion to 2.8 trillion parameters, including NVIDIA’s Nemotron 3 Ultra, OpenAI’s gpt-oss-120b, and Google’s Gemma 4, along with embedding models for retrieval-augmented generation. Akamai Cloud connects the service to managed databases and object storage, so models can draw on enterprise data, while Akamai handles orchestration, model execution, and authentication.
A bounded place for agents to run code
Akamai Agent Sandbox is the security half of the launch. Agents run on Akamai Functions, and the code they write runs in its own isolated environment, separate from the agent’s runtime. Delivered as an open-source component of the Spin framework, the sandbox uses WebAssembly isolation to set clear limits on compute, memory, outbound connections, and system capabilities. Akamai said it also plans to add capabilities that govern what agents can reach — and what can reach them.
More GPUs underneath
The services sit on an expanding hardware fleet. Akamai said it has added NVIDIA Blackwell Ultra GPU clusters alongside its RTX PRO 6000 Blackwell Server Edition GPUs; each eight-GPU Blackwell Ultra node carries 288 GB of HBM3e memory per GPU, 2.3 TB in total. The company is also preparing its data centers to support NVIDIA’s Vera Rubin platform in 2027. Akamai said the managed services build on an infrastructure foundation that has already secured multiyear Akamai Cloud contracts collectively worth more than $14 billion.
Why it matters
The agent era has a plumbing problem: models are easy to rent, but serving them close to users, grounding them in company data, and fencing in the code they write is still a do-it-yourself project at most companies. Akamai’s bet is that enterprises will pay for that plumbing from the same network that already delivers and secures their traffic. For Tested Intelligence readers, this is the infrastructure layer of the AI stack quietly becoming a product — and another sign that “where does the agent run, and what can it touch?” is now a board-level question, not an engineering detail.
Frequently asked questions
What is Akamai Inference?
A managed model-serving service on Akamai Cloud. It offers 12 open-weight models at launch, from 8 billion to 2.8 trillion parameters, through OpenAI-compatible APIs, in serverless pay-per-use or dedicated deployments.
What is Akamai Agent Sandbox?
An isolated execution environment for AI agents, built on Akamai Functions and the open-source Spin framework. Code an agent writes runs in a WebAssembly sandbox with enforced limits on compute, memory, outbound connections, and system capabilities.
Are the new Akamai AI services generally available?
No. Akamai said both Akamai Inference and Akamai Agent Sandbox are available in beta, with the Inference beta including serverless deployment options.
Sources: GlobeNewswire (Akamai), SiliconANGLE