TL;DR — Key Takeaways
-
Akamai brings AI closer to the edge. New managed inference services give enterprises access to open-weight models across Akamai’s distributed infrastructure.
-
AI agents get a secure sandbox. Akamai Agent Sandbox uses WebAssembly and serverless technology to isolate and run AI agents in a managed environment.
-
Enterprises gain more AI control. Open-weight models and managed infrastructure could help reduce costs, improve data control and limit dependence on proprietary AI providers.
Akamai today added a managed sandbox service for deploying artificial intelligence (AI) agents, along with an additional managed inference service to provide access to open-weight and custom AI models that are both available now in beta.
Initially, the Akamai Inference service provides access to 12 open-weight models ranging from 8 billion to 2.8 trillion parameters that are accessed via an OpenAI-compatible application programming interface (API). The catalog includes NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b, and Google Gemma 4 model and a set of embedded models for retrieval-augmented generation (RAG) use cases. That service complements an editing set of managed databases and storage services that Akamai already provides.
The Akamai Agent Sandbox, meanwhile, provides a way to deploy AI agents using a serverless Akamai Functions framework in a secure runtime environment. Based on the WebAssembly framework that Akamai gained with the acquisition of Fermyon, Agent Sandbox is an open-source component of the Spin framework that is being advanced under the auspices of the Cloud Native Computing Foundation (CNCF).
Via those managed services, IT teams can access more than 4,350 points of presence where Akamai has deployed clusters based on a mix of traditional CPUs and graphical processing units (GPUs) from NVIDIA to minimize network latency. Akamai is also preparing to add support for NVIDIA Vera Rubin processors in 2027.
As organizations begin to deploy their own custom AI models, demand for managed services is increasing, says Ari Weil, vice president of product marketing and cloud evangelist for Akamai. Organizations want to reduce their dependency on providers of frontier AI model services to reduce costs, he adds. A managed service provides access to open-weight AI models to reduce costs in a way that reduces the amount of internal expertise required to successfully deploy them, noted Weil.
The Akamai Inference service also gives IT teams the ability to deploy AI models in the cloud or at the network edge where the output of AI models is being consumed and, increasingly, created as real-time applications are deployed, he added.
It’s not clear just how rapidly organizations are adopting open-weight models as alternatives to frontier AI models, but as costs continue to rise, many of them will be looking for ways to reduce their total costs. Deploying an AI model themselves eliminates the input and output token costs that are created when relying on an external service versus self-hosting an AI model that is invoked via a catalog. In general, an open-weight model is sufficiently powerful to automate most of the tasks of the average enterprise.
There may, of course, always be use cases where the capabilities of a more advanced AI model are required, but as organizations become more adept at building and deploying AI applications and agents, many of them will increasingly rely on AI models they have more control over to both reduce costs and ensure security. Many organizations are especially sensitive about sharing intellectual property with providers of AI model services that might be used to train the next iteration of that model. The concern is that at some point in the future a simple prompt might generate a proprietary trade secret.
Regardless of approach, the one thing that is certain is that the AI ecosystem is becoming more diverse with each passing day. The challenge now is figuring out how to manage and secure it.
