Akamai launches AI inference and agent sandbox services on Akamai Cloud

Akamai Technologies has introduced two services, Akamai Inference and Akamai Agent Sandbox, to help enterprises deploy AI models and run AI agents on its cloud platform. The services are currently available in beta and are designed to address model deployment, data access and the secure execution of code generated by AI agents.

Announced on October 28, 2026, the services add managed AI inference and isolated execution environments to Akamai Cloud. The company said the offerings would allow enterprises to access open-weight and custom AI models, connect them to their own data, and run AI agents within controlled environments.

Akamai Inference provides managed access to AI models through application programming interfaces (APIs), while Akamai Agent Sandbox provides an isolated environment for executing code written by AI agents. The two services form part of Akamai’s broader strategy to expand its cloud infrastructure and services for AI workloads.

The company said enterprises could use the services alongside its existing compute, storage, networking and security capabilities. Its infrastructure spans central cloud regions and more than 4,350 edge locations.

According to Akamai, its cloud business has secured multiyear contracts collectively worth more than $14 billion. The announcement did not specify how much of this contracted value relates to AI services.

Managed access to AI models

Akamai Inference allows enterprises to access open-weight and custom AI models without having to manage the underlying model-serving infrastructure themselves.

The service supports OpenAI-compatible APIs, which the company said would allow applications already using compatible interfaces to connect to the service without requiring code rewrites. Customers can choose between serverless inference, billed on a pay-per-use basis, and dedicated deployments.

At launch, the model catalogue includes 12 open-weight models, ranging from 8 billion to 2.8 trillion parameters. These include NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4. The service also supports embedding models used in retrieval-augmented generation (RAG), a technique that enables AI systems to retrieve relevant information from external data sources when generating responses.

Akamai Inference can connect with managed databases and object storage available through Akamai Cloud. It can also connect to private and public cloud platforms, allowing enterprises to use their existing data with the models.

The service manages model orchestration, execution and authentication. Requests are processed through Akamai’s global network, which the company said is intended to reduce latency and support service availability and performance.

The offering targets organisations that want to deploy AI-powered applications without building and maintaining the entire model-serving stack themselves.

Sandbox service isolates AI-generated code

Akamai Agent Sandbox focuses on the execution risks associated with AI agents. As these systems increasingly perform tasks and generate code, enterprises need mechanisms to restrict what that code can access and execute.

The service provides an environment in which AI agents can run while the code they generate executes in a separate, isolated environment. According to Akamai, agents run on Akamai Functions, with the generated code isolated from the agent’s own runtime.

The sandbox uses WebAssembly-based isolation and is being delivered as an open-source component of the Spin framework. It is designed to impose limits on computing resources, memory, outbound network connections and system capabilities.

These restrictions are intended to limit the resources and systems that AI-generated code can access during execution.

Akamai said developers could choose from different infrastructure options depending on their workloads, including dedicated CPU infrastructure, NVIDIA-accelerated systems, managed Kubernetes and Akamai Functions for serverless execution across core and edge locations.

The company also plans to introduce additional capabilities to govern which resources AI agents can access and which entities can interact with them. It did not provide a timeline for these planned additions.

Expansion of GPU infrastructure

Alongside the two services, Akamai announced an expansion of its dedicated graphics processing unit (GPU) infrastructure to include NVIDIA Blackwell Ultra GPU clusters. The company is also preparing its data centres to support NVIDIA’s Vera Rubin platform in 2027.

Its existing GPU infrastructure includes NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, which Akamai positions for low-latency inference workloads. The Blackwell Ultra clusters are intended for enterprises that want to develop, fine-tune and host models on dedicated infrastructure.

According to Akamai, each eight-GPU Blackwell Ultra node has 288 GB of HBM3e memory per GPU, amounting to 2.3 TB across the eight GPUs. The infrastructure is designed to support memory-intensive workloads, including extended reasoning, test-time scaling and AI agents handling long-context tasks.

Dedicated GPU deployments also give enterprises greater control over model versions and upgrade schedules. Akamai said these deployments would operate without rate limits or queueing associated with shared-tenancy environments.

The GPU infrastructure connects with the company’s existing data, networking and security services.

Jon Alexander, senior vice-president of cloud technology at Akamai, said the platform was intended to support organisations with different deployment requirements, from managed AI services to dedicated GPU infrastructure and isolated environments for AI-generated code.

Beta availability

Akamai Inference and Akamai Agent Sandbox are available in beta. The inference service’s beta includes serverless deployment options, while the Agent Sandbox is accessible through the Spin software development kit (SDK).

Enterprises interested in NVIDIA B300 GPU clusters have been asked to contact Akamai’s sales team for availability details.

The launch expands Akamai Cloud’s AI capabilities across three areas: managed model inference, isolated execution for AI agents and dedicated GPU infrastructure for developing and running AI workloads. The company plans to build on these capabilities with further controls over agent access and additional infrastructure support.