Extreme Networks to launch on-premises AI zones using idle network hardware

Tapping into idle hardware

Extreme Networks is preparing to launch a new capability called Exchange zones. This technology is designed to help organizations run AI applications locally on their own network infrastructure instead of routing all data to the cloud. The company plans to make this available during the first half of 2027.

The core idea is based on the fact that many enterprise networks contain significant amounts of idle compute capacity. These are spare processing resources sitting in hardware that is already deployed at various locations, such as retail stores, warehouses, and branch offices. Extreme Networks wants to put that unused power to work for AI tasks.

By using the hardware customers already own, the company aims to provide a more efficient way to handle AI workloads. This approach addresses the growing demand for better cost control and operational resilience. It allows businesses to manage their AI needs without relying entirely on external cloud providers.

How Exchange zones function

An Exchange zone combines an Extreme Universal Compute Platform or a qualifying server with spare capacity from existing network devices. The system is designed to automatically distribute workloads based on specific rules set by the user. These rules can include requirements for how fast the data needs to move, where it is allowed to stay, and how resilient the connection must be.

Under this architecture, AI models are designed to run locally by default. The system only sends data to the cloud for inference if the customer's specific policy allows it for a particular task. If any data does leave the local boundary for cloud processing, the system keeps a detailed record of that exception.

This setup provides a high level of observability for the user. It tracks what stayed local, which models were used, and the lifecycle of the data. This level of tracking is intended to help organizations maintain a clear record of their AI operations while still benefiting from the flexibility of the cloud when necessary.

Solving for data and cost

The company argues that cloud-only architectures are poorly suited for several common enterprise use cases. Some applications cannot handle the delay, or latency, caused by sending data back and forth to a remote cloud server. Other applications generate such large volumes of data that moving and storing it remotely becomes too expensive.

Exchange zones specifically target three areas: data sovereignty, survivability, and cost. For data sovereignty, the zones keep AI processing within a single site and a single country. This is a major requirement for regulated industries that must keep sensitive information within specific borders.

For survivability, the zones ensure that time-sensitive AI applications keep running even if the internet connection fails. This is critical for maintaining operations at distributed sites like warehouses. Finally, the approach reduces costs by minimizing the amount of data that needs to be moved and processed by expensive cloud providers.

The broader AI stack

Exchange zones are one part of a larger five-layer enterprise stack from Extreme Networks. The foundation of this stack is the network infrastructure where data is generated and where the compute capacity currently lives. Above that sits Extreme Platform ONE, which handles unified management and security for the network.

The intelligence layer of the stack is called Extreme AI, while Extreme Agent ONE serves as the interface for users to interact with the AI. Exchange zones act as the extension that brings these AI skills and applications to the edge of the network. This positions the company to compete as more businesses question if all their AI needs to be in the cloud.

As data sovereignty regulations get stricter around the world, the demand for local processing is expected to grow. Extreme Networks is positioning its products to meet these needs as the costs of cloud-based inference continue to add up for large companies. They plan to share more details about the specific capabilities of Exchange zones as the 2027 launch date approaches.