Akamai Debuts AI Inference, Agent Sandbox

Akamai Technologies on Tuesday, Oct. 27, launched Akamai Inference and Akamai Agent Sandbox, two beta services that let companies serve open-weight AI models and run AI agents inside enforced limits on its cloud. The Cambridge, Massachusetts, company best known for content delivery says its AI infrastructure has already secured multiyear Akamai Cloud contracts collectively worth more than $14 billion.

The launch pairs a managed model-serving layer with an isolated place to execute the code agents write. Akamai did not publish prices for either service.

Key takeaways

• Akamai Inference serves 12 open-weight models at launch, from 8 billion to 2.8 trillion parameters, through OpenAI-compatible APIs, with serverless pay-per-use or dedicated options.
• Akamai Agent Sandbox runs agent-generated code apart from the agent itself, using WebAssembly isolation from the open-source Spin framework to cap compute, memory, outbound connections and system access.
• Both services are in beta. Akamai gave no pricing and says it is preparing its data centers for Nvidia's Vera Rubin platform in 2027.

What is Akamai Inference?

Akamai Inference is a managed service for open-weight and custom models. According to Akamai's announcement, the launch catalog includes Nvidia Nemotron 3 Ultra, OpenAI's gpt-oss-120b and Google's Gemma 4, plus embedding models for retrieval-augmented generation (RAG).

Because the APIs are OpenAI-compatible, Akamai says existing applications can move over without rewrites. The service connects to managed databases, object storage and private or public clouds, while Akamai handles orchestration, model execution and authentication. Requests run on its global network of core regions and more than 4,350 edge locations.

How Akamai Agent Sandbox isolates agent code

Agent Sandbox runs agents on Akamai Functions, the company's serverless platform, and executes the code they generate in an environment separate from the agent's own runtime. It ships as an open-source component of Spin, the WebAssembly framework Akamai gained when it announced its acquisition of Fermyon on Dec. 1, 2025. Fermyon maintains the Spin and SpinKube projects at the Cloud Native Computing Foundation.

Customers can choose among microVMs with snapshot and fork, policy-enforced containers and WebAssembly-isolated functions. "Every team running agents needs a boundary around the code those agents write," said Jon Alexander, Akamai's senior vice president of cloud technology. Akamai says it will add controls over what agents can reach and what can reach them.

Akamai's Nvidia GPU lineup

For dedicated capacity, Akamai offers Nvidia RTX PRO 6000 Blackwell Server Edition GPUs for low-latency inference and Blackwell Ultra clusters for building, tuning and self-hosting models. Each eight-GPU Blackwell Ultra node carries 288 GB of HBM3e memory per GPU, or 2.3 TB in total. Akamai says those clusters come with no rate limits or shared-tenancy queueing, and asks customers to contact sales about B300 cluster availability.

Why Akamai is betting on inference

Adam Karon, chief operating officer and general manager of Akamai Cloud, framed the launch around customers consolidating their AI stack. "They want to build, run, and protect AI on the same platform," he said. SiliconANGLE reported that Karon expects inference to become the main AI focus for most enterprises.

Among single customers, Anthropic committed $11.6 billion over seven years to Akamai Cloud in September. Isolation has also become a selling point as agents gain autonomy; researchers have documented agents that turned to SQL injection when ordinary requests failed.

What's next

Akamai has not said when either service will leave beta or how it will be priced. The company is readying its data centers for Nvidia Vera Rubin in 2027. Independent benchmarks of the latency and isolation claims have yet to appear.

FAQ

Can developers use Akamai Inference today?

Yes, in beta. The beta includes serverless deployment options, and the APIs follow OpenAI's format. Akamai has not published pricing.

How is Agent Sandbox different from running agent code in a container?

Agent Sandbox separates the code an agent writes from the agent itself. Customers can pick WebAssembly isolation, policy-enforced containers or microVMs with snapshot and fork, with limits on compute, memory and outbound connections.

References

• Akamai Launches AI Inference Service and Agent Execution Environment — Akamai (Oct. 27, 2026)
• Akamai Technologies Announces Acquisition of Function-as-a-Service Company Fermyon — Akamai (Dec. 1, 2025)
• Akamai expands its AI Cloud platform with dedicated inference and agent execution environments — SiliconANGLE (Oct. 27, 2026)