The cloud and cybersecurity company is expanding into managed AI model deployment and isolated agent execution, challenging how enterprises distribute computing infrastructure.
Akamai Technologies has introduced two artificial intelligence infrastructure services designed to help enterprises deploy AI models and run autonomous agents within controlled execution environments.
The October 27 announcement introduced Akamai Inference and Akamai Agent Sandbox, expanding the company’s cloud platform beyond traditional content delivery, security and computing services.
Both products are initially available in beta.
The launch reflects growing demand for infrastructure that supports AI applications closer to users while allowing developers to control how autonomous systems access computing resources and business data.
Managed AI Inference Expands Model Deployment Options
Akamai Inference provides managed access to open-weight and custom AI models through OpenAI-compatible application programming interfaces.
This allows developers to connect existing applications without completely rebuilding their model integrations.
According to Akamai’s product announcement, the initial catalog contains 12 open-weight models ranging from 8 billion to 2.8 trillion parameters.
Available options include NVIDIA Nemotron 3 Ultra, OpenAI’s gpt-oss-120b and Google’s Gemma 4, alongside embedding models supporting retrieval-augmented generation.
Customers can choose between serverless, usage-based inference and dedicated deployments.
Akamai also integrates model serving with managed databases, object storage and network infrastructure.
Agent Sandbox Introduces Isolated Execution
The second product addresses a different technical challenge: autonomous AI agents that write and execute their own code.
Akamai Agent Sandbox provides an isolated environment intended to prevent agent-generated programs from freely accessing the surrounding infrastructure.
The company uses WebAssembly-based isolation, delivered through the open-source Spin framework, to establish restrictions on computing resources, memory, outbound connections and system capabilities.
Agents operate through Akamai Functions, while generated code executes separately from the agent runtime.
The approach aims to reduce the consequences of unintended or unauthorized agent activity.
However, sandboxing does not eliminate every potential security risk, particularly those involving overly broad permissions or legitimate tools used improperly.
Akamai Expands Its Computing Infrastructure
The new services build on Akamai’s global network, which spans more than 4,350 edge locations.
The company said its cloud infrastructure has secured multiyear contracts collectively exceeding $14 billion.
That figure represents contracted business rather than revenue recognized from the newly launched products.
Akamai has also expanded its GPU infrastructure with NVIDIA Blackwell Ultra clusters.
Each eight-GPU node contains approximately 2.3 terabytes of high-bandwidth memory, supporting demanding model inference and long-context workloads.
The company plans to support NVIDIA’s Vera Rubin platform in 2027.
Adam Karon, Akamai’s chief operating officer and general manager of Akamai Cloud, said enterprise demand increasingly centers on running AI applications, models and agents through a connected infrastructure platform.
The Competition Is Moving Beyond Model Performance
For developers, the launch highlights how AI infrastructure competition is expanding beyond access to powerful models.
Companies must now evaluate inference latency, deployment costs, data residency, execution isolation and interoperability.
In an October 27 analysis, analyst Chris DePuy of 650 Group identified Akamai’s distributed network and established cybersecurity business as potential advantages over providers operating primarily through concentrated cloud regions.
However, Akamai must still demonstrate the commercial performance, reliability and security effectiveness of its beta services.
The company’s strategy reflects a broader industry shift toward infrastructure that combines model execution, application security and operational control.
For enterprise AI builders, choosing where an AI model runs may increasingly become as consequential as choosing which model to use.
