New beta services bring managed model serving and isolated AI agent execution to Akamai Cloud, alongside expanded GPU infrastructure for workloads spanning core data centres and edge locations.
Akamai has introduced two services designed to help enterprises move AI applications from experimentation into production, addressing challenges around inference costs, latency and the security of increasingly autonomous AI agents.
The new offerings, Akamai Inference and Akamai Agent Sandbox, extend Akamai Cloud with managed access to AI models and isolated environments for executing agent-generated code. The company is positioning the services as part of an integrated platform that combines computing, data access, networking and security across core infrastructure and more than 4,350 edge locations.
As enterprises deploy AI at scale, the operational challenge is shifting beyond model training. Organizations must serve models efficiently, connect them to business data and control what AI agents can access when they perform tasks or execute code. Akamai aims to bring these capabilities together rather than requiring customers to assemble and secure multiple services independently.
Managed model serving without infrastructure overhead
Akamai Inference provides access to open-weight and custom models through OpenAI-compatible APIs, allowing existing applications to connect without requiring rewrites. Customers can select serverless, pay-per-use inference or dedicated deployments.
The launch catalogue includes 12 open-weight models, ranging from 8 billion to 2.8 trillion parameters, as well as embedding models used in retrieval-augmented generation (RAG). The service also connects models to managed databases, object storage and public or private cloud platforms, while handling orchestration, execution and authentication.
“Enterprises should not have to choose between speed and control when they put AI into production.” — Jon Alexander, Senior Vice President of Cloud Technology, Akamai
Akamai says its distributed network can help reduce latency and support availability and consistent performance. These capabilities could be particularly relevant for applications requiring frequent inference requests or responses close to end users and enterprise data.
Isolating AI agents and their code
Akamai Agent Sandbox addresses a different risk: agents that write and execute their own code. The service runs agents on Akamai Functions while placing the code they generate in a separate, isolated environment.
Delivered as an open-source component of the Spin framework, the sandbox uses WebAssembly isolation to impose limits on computing resources, memory, outbound connections and system capabilities. This separation is intended to restrict what agent-generated code can reach if it behaves unexpectedly.
The company also plans to introduce further capabilities governing what agents can access and which systems can interact with them.
Expanded GPU infrastructure
Alongside the new services, Akamai has expanded its dedicated GPU fleet to include NVIDIA Blackwell Ultra clusters. It is preparing its data centres for the NVIDIA Vera Rubin platform in 2027.
The infrastructure also includes NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for low-latency inference. Dedicated GPU clusters are aimed at organizations that want to build, fine-tune and host models on infrastructure they control, with greater autonomy over model versions and deployment schedules.
Akamai Inference and Akamai Agent Sandbox are currently available in beta. The Inference beta includes serverless deployment options, while availability of NVIDIA B300 GPU clusters is subject to confirmation from Akamai.
For enterprises, the significance of the launch lies in the combination of model serving and agent execution controls. As AI systems take on more operational responsibilities, the ability to manage inference performance while enforcing boundaries around autonomous actions will become an increasingly important part of enterprise AI infrastructure.