How to run and secure AI agents at scale: Akamai Inference and Agent Sandbox

Enterprises moving AI from pilot to production face a fragmented stack: serving infrastructure, data access, routing, governance and a hard boundary around code that agents write and run. Assembling these pieces from separate providers increases latency, cost and attack surface. Akamai now consolidates them on a single global platform spanning core GPU regions and more than 4,350 edge locations. Softprom is the official distributor of Akamai.

What was announced

Akamai (Nasdaq: AKAM) launched two foundational products for its AI stack: Akamai Inference and Akamai Agent Sandbox. The company also expanded its dedicated GPU fleet with NVIDIA Blackwell Ultra clusters and is preparing data centers for the NVIDIA Vera Rubin platform in 2027. Akamai Cloud has already secured multiyear contracts collectively worth more than $14 billion.

Akamai Inference provides managed access to open-weight and custom models through OpenAI-compatible APIs. The launch catalog includes 12 open-weight models from 8 billion to 2.8 trillion parameters, among them NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4, plus embedding models for retrieval-augmented generation. Customers choose serverless pay-per-use or dedicated deployments.

Akamai Agent Sandbox, delivered as an open source component of the Spin framework, gives developers an isolated environment for agents and the code they generate, using WebAssembly isolation to enforce limits on compute, memory, outbound connections and system capabilities.

Akamai Inference and Akamai Agent Sandbox add the inference and execution layers that turn our AI infrastructure into a full AI stack, and our security portfolio protects every layer of it

Why this matters

For CIOs, CISOs and procurement leaders, the shift to agentic AI introduces new risk vectors: autonomous code execution, uncontrolled outbound calls and shared-tenancy performance variability. A connected platform that unifies inference, agent isolation, data residency and security reduces integration overhead and vendor sprawl. Dedicated GPU clusters remove rate limits and queueing, giving teams control over model versions and deprecation timelines.

Technical details

• Inference API: OpenAI-compatible, serverless or dedicated.
• Model catalog: 12 open-weight models, 8B to 2.8T parameters.
• GPU fleet: NVIDIA RTX PRO 6000 Blackwell Server Edition and Blackwell Ultra clusters.
• Blackwell Ultra node: 8 GPUs, 288 GB HBM3e per GPU, 2.3 TB total.
• Agent isolation: WebAssembly via Spin framework (open source).
• Edge footprint: 4,350+ edge locations connected to core regions.
• Availability: Both products in beta; NVIDIA B300 GPU clusters via Akamai Sales.

FAQ

What is Akamai Inference?

It is a managed service for running open-weight and custom AI models through OpenAI-compatible APIs on Akamai Cloud, with serverless or dedicated deployment options and 12 models at launch.

How does Akamai Agent Sandbox isolate agent code?

It uses WebAssembly isolation delivered via the open source Spin framework, enforcing limits on compute, memory, outbound connections and system capabilities, so generated code runs separately from the agent runtime.

Which GPUs are available for self-hosted model tuning?

NVIDIA RTX PRO 6000 Blackwell Server Edition for low-latency inference and NVIDIA Blackwell Ultra clusters for building and tuning frontier-class models, with Vera Rubin planned for 2027.

Softprom and Akamai

Softprom is the official distributor of Akamai. Our team supports enterprises in evaluating, deploying and scaling Akamai Cloud, security and AI services.

Prepared by the Softprom editorial team based on vendor materials (Softprom DistriFlow). Source: original article.