Akamai has quietly become one of the biggest names in AI infrastructure — and on October 27, 2026, it took its biggest step yet toward a complete AI stack. The company launched Akamai Inference and Akamai Agent Sandbox on Akamai Cloud, two products designed to help enterprises run models and AI agents in production, not just in demos.
What Akamai Inference actually does
Akamai Inference gives companies managed access to open-weight and custom models through OpenAI-compatible APIs. That detail matters: existing applications can move over without being rewritten. Customers can choose serverless, pay-per-use inference or dedicated deployments, and the launch catalog includes 12 open-weight models ranging from 8 billion to 2.8 trillion parameters. The named models include NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4, along with embedding models for retrieval-augmented generation — the technique that lets a model pull in a company's own documents before answering. Akamai also provides managed databases and object storage, so models can draw on enterprise data, and it handles orchestration, execution and authentication.
Why agents need a sandbox
AI agents are increasingly doing real work: taking actions on a user's behalf and writing and running their own code. That is powerful, and risky. Akamai Agent Sandbox gives developers a secure place to run both the agent and the code it writes, with the code running in its own isolated environment, separated from the agent's runtime. Delivered as an open-source component of the Spin framework, the sandbox uses WebAssembly isolation to set clear limits on compute, memory, outbound connections and system capabilities. In plain terms, the agent can work, but it cannot wander. That idea connects directly to the wider agent wave: computer-using agents built to click, type and fill in forms inside everyday apps — as in OpenAI's GPT-6.1 Terra — only make sense if their code runs inside a boundary.
The edge advantage: 4,350 locations and Blackwell Ultra
Akamai's pitch is distribution. Its cloud now spans dense GPU and CPU clusters in core regions plus more than 4,350 edge locations, protected by the same security portfolio that defends many of the world's largest online businesses. Running inference close to users cuts latency — the delay between asking and answering — which matters for real-time apps. On hardware, Akamai expanded its dedicated GPU fleet to include NVIDIA Blackwell Ultra clusters and is preparing its infrastructure to support NVIDIA's Vera Rubin platform in 2027. Each eight-GPU Blackwell Ultra node carries 288 GB of HBM3e memory per GPU, 2.3 TB in total, which Akamai says is enough to hold frontier-class models. The company also offers RTX PRO 6000 Blackwell Server Edition GPUs for low-latency inference.
The money behind the move
Akamai says the foundation underneath these products has already secured multiyear contracts for Akamai Cloud collectively worth more than $14 billion. That figure, reported by the company, is a sign of how much enterprises are now spending on AI infrastructure rather than on models alone. It fits a pattern across the industry this month. Meta's funding blitz worth up to $95 billion shows one route to paying for AI expansion, while SpaceX leasing about $40 billion worth of AI chips shows a very different one. Different strategies, same message: the bill for running AI at scale is now measured in tens of billions, and whoever controls the compute controls the pace.
What is in beta, and what to watch
Both Akamai Inference and Akamai Agent Sandbox are available in beta, Akamai said, with serverless deployment options for Inference. Contact Akamai Sales for NVIDIA B300 GPU cluster availability. As with any beta, early customers should expect rough edges — and the real test will be whether developers actually build on it, and whether the sandbox model becomes a standard way to contain agent code. The model side is also moving fast in parallel: Anthropic's record-breaking IPO filing is a reminder that the companies building frontier models are themselves scaling up just as the infrastructure to run them is maturing. Akamai is betting that the next bottleneck is not the model, but where and how safely it runs.
The bottom line
Akamai is not launching a chatbot. It is launching the plumbing: serve the model, run the agent, keep both inside a boundary, and do it everywhere from a core data center to an edge site near the user. If agentic AI is going to move from demos to daily work, that plumbing has to work — and Akamai wants to be the company that provides it. Would you let an AI agent run code on your systems if it was locked inside a sandbox? Tell us in the comments.
Comments
Post a Comment