Akamai Launches AI Inference and Agent Sandbox on Akamai Cloud
October 30, 2026 · 2 min read
Akamai has launched Akamai Inference, a managed model serving service, and Akamai Agent Sandbox, an isolated environment for running AI agent code on Akamai Cloud.
Akamai Inference allows organizations to access open weight models and customized models via an API compatible with OpenAI. At launch, 12 open weight models are available, ranging from 8 billion to 2.8 trillion parameters, including Google Gemma 4, OpenAI gpt-oss-120b, and NVIDIA Nemotron 3 Ultra. The service also provides embedding models for Retrieval Augmented Generation. Customers can choose between serverless billing based on usage or dedicated deployment.
Akamai Agent Sandbox provides a space to run AI agents and the code they generate. The agents operate on Akamai Functions, while the generated code runs in an environment separate from the agent runtime. This sandbox is an open source component of the Spin framework and utilizes WebAssembly isolation to define limits for memory, compute, system capability, and outbound connections.
To support these services, Akamai is expanding its GPU fleet with NVIDIA Blackwell Ultra. Each Blackwell Ultra node contains 8 GPUs with 288 GB of HBM3e memory per GPU, totaling 2.3 TB. This infrastructure is intended for frontier models and workloads involving agentic tasks, chain of thought reasoning, and test time scaling. Akamai also offers NVIDIA RTX PRO 6000 Blackwell Server Edition for low latency inference.
Akamai Cloud provides compute resources through CPU and GPU clusters in core regions and over 4,350 edge locations. The company has secured multi year contracts for Akamai Cloud valued at over 14 billion dollars. Akamai is also preparing data centers to support the NVIDIA Vera Rubin platform in 2027.
Both Akamai Inference and Akamai Agent Sandbox are currently available in beta. The beta for Akamai Inference includes serverless deployment options.