Akamai Launches AI Inference and Agent Sandbox for Enterprise AI

Akamai has launched two new products that turn its cloud platform into a full stack for enterprise artificial intelligence. Announced on October 27, 2026, from Cambridge, Massachusetts, Akamai Inference and Akamai Agent Sandbox are aimed at companies that want to run AI models and AI agents without stitching together services from several different providers.

The launch reflects a shift in what businesses actually need from AI. Building or choosing a model is only the first step. The harder work is serving that model reliably, connecting it to company data, and controlling what an autonomous agent can do once it starts writing and running its own code. Akamai says its new products address both sides of that problem on a single platform, protected by the same security portfolio it uses for many of the world’s largest online businesses.

What Akamai Inference Does

Akamai Inference gives enterprises managed access to open-weight and custom models through OpenAI-compatible APIs, so existing applications can move over without rewrites. Customers can choose serverless, pay-per-use inference or dedicated deployments, depending on how much control and capacity they need.

At launch, the catalog includes 12 open-weight models ranging from 8 billion to 2.8 trillion parameters, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b, and Google Gemma 4, along with embedding models for retrieval-augmented generation. That range matters because different jobs need different models: a small model can answer routine questions cheaply, while a frontier-class model is needed for complex reasoning.

Akamai Cloud also provides managed databases and object storage, and connects to private and public cloud platforms, so models can draw on enterprise data. Akamai Inference then handles orchestration, model execution, and authentication, while requests run on the Akamai global network to reduce latency and keep performance consistent. The company says the underlying infrastructure has already secured multiyear contracts for Akamai Cloud collectively worth more than $14 billion.

A Sandbox for Agents That Write Code

AI agents are increasingly doing real work, from taking actions on a user’s behalf to writing and running their own code. That creates a new risk: code written by an agent should never have unlimited access to the system it runs on. Akamai Agent Sandbox is designed to solve that problem by giving developers a secure place to run both the agent and the code it produces.

Agents run on Akamai Functions, while the code they write runs in its own isolated environment, sequestered from the agent’s runtime. Delivered as an open source component of the Spin framework, Akamai Agent Sandbox uses WebAssembly isolation to set clear limits on compute, memory, outbound connections, and system capabilities. Developers can therefore deploy agents with confidence about where they run and what they can reach.

Akamai notes that agent isolation is converging on several approaches, from microVMs with snapshot and fork capability to policy-enforced containers and WebAssembly-isolated functions. Its platform lets developers choose the architecture that fits their workload, with NVIDIA-accelerated and dedicated CPU infrastructure, managed Kubernetes, and Akamai Functions for serverless execution across edge and core. The company also plans to add capabilities that govern what agents can reach and what can reach them.

The hard part of enterprise AI is no longer building a model, but running it safely at scale.

The hard part of enterprise AI is no longer building a model, but running it safely at scale. That is the gap Akamai is targeting: one platform for serving models, running agents, and securing both, instead of a patchwork of tools that each need their own security review.

Bigger GPUs: Blackwell Ultra Today, Vera Rubin Next

Alongside the software launch, Akamai expanded its dedicated GPU fleet to include NVIDIA Blackwell Ultra GPU clusters and is preparing its infrastructure to support the NVIDIA Vera Rubin platform next year. Its GPU fleet now spans NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for low-latency inference and Blackwell Ultra clusters for enterprises that want to build, tune, and self-host models on infrastructure they control.

Each eight-GPU Blackwell Ultra node carries 288 GB of HBM3e memory per GPU, 2.3 TB in total, enough to hold frontier-class models. The Blackwell Ultra architecture targets the workloads driving that memory demand: test-time scaling, chain-of-thought reasoning, and long-context agentic work. GPU clusters give enterprises control over model versions and deprecation timelines, and run without rate limits or shared-tenancy queueing. They connect to the same data, networking, and security services as the rest of Akamai Cloud.

That combination is also relevant to the wider AI infrastructure race. Chip and cloud capacity has become a strategic bottleneck, as shown by NVIDIA’s major AI partnership moves and by new storage breakthroughs built for AI workloads. Akamai is betting that customers will prefer a platform that covers compute, data, networking, and security together, rather than assembling each layer separately.

Why This Matters for Enterprise AI

For most companies, the choice has been between speed and control. A fully managed service is fast to start but limits customization, while self-hosted infrastructure offers control at the cost of complexity. Akamai’s pitch is that both options can live on one platform: teams that want to ship an AI feature quickly can use the managed service, while teams that need dedicated GPUs for frontier-class open models can run them on their own terms.

Security is the other half of the argument. As agents gain the ability to act, the boundary around their code becomes as important as the model itself. A sandbox that limits what agent-written code can touch is becoming a basic requirement, not a premium feature, especially in regulated industries. The same concern is driving new safety work elsewhere in AI, including the UK’s AI biosecurity lab work with leading AI developers.

• Two new products: Akamai Inference for managed model serving and Akamai Agent Sandbox for isolated agent execution.
• 12 open-weight models at launch, from 8 billion to 2.8 trillion parameters, via OpenAI-compatible APIs.
• More than 4,350 edge locations plus core GPU and CPU clusters on one connected platform.
• WebAssembly isolation limits agent code on compute, memory, outbound connections, and system capabilities.
• Blackwell Ultra GPU clusters available now; Vera Rubin platform support being prepared for 2027.

Akamai chief operating officer Adam Karon said the company had been implementing an AI infrastructure strategy based on the idea that inference would represent the majority of enterprise focus and investment, and that customers had pulled the company further in that direction. Jon Alexander, senior vice president of cloud technology at Akamai, said enterprises should not have to choose between speed and control when they put AI into production.

What to Watch Next

Akamai Agent Sandbox, via the Spin SDK, and Akamai Inference are available in beta, with serverless deployment options included in the Inference beta. Availability of NVIDIA B300 GPU clusters is through Akamai Sales. The next things to watch are how the beta products perform in production, what governance capabilities Akamai adds for controlling what agents can reach, and how its preparation for the NVIDIA Vera Rubin platform in 2027 progresses. Those steps will show whether a single-platform approach to inference and agent execution can win over enterprises that currently assemble their AI stacks piece by piece.