Extreme Networks previews Exchange zones for on-premises AI

Extreme Networks has previewed Extreme Exchange zones, a planned on-premises AI offering that combines local servers with spare capacity in the company’s installed access points and switches. General availability is planned for the first half of 2027, and the company has not disclosed which hardware will qualify or how it will determine available processing capacity.

The announcement separates two parts of Extreme’s AI strategy. Its Platform ONE and Agent ONE products use AI to help operate networks. Exchange zones are intended to turn network infrastructure into a local runtime for other AI applications, agents and plug-ins at branches, retail sites, warehouses and other distributed locations.

How would Extreme Exchange zones work?

Under Extreme’s proposed design, a zone would combine an Extreme Universal Compute Platform or a qualifying customer-owned server with unused compute capacity from deployed Extreme access points and switches. The company says the system would place workloads according to customer-defined requirements for latency, data residency and resilience.

Models would run locally by default, according to Extreme. Cloud inference would be permitted only when a customer’s policy allows it, with each such use recorded. Extreme says each zone would be tied to one site in one country and provide records on data residency, model use and data life-cycle management.

The architecture is aimed at workloads where moving data to cloud infrastructure may create residency, latency or connectivity concerns. Cloud infrastructure remains appropriate for many AI jobs, but Extreme argues that locally generated, high-volume or sensitive data can be costly or impractical to send off-site. Those are design goals and company claims, not demonstrated results from an available product.

What is new compared with Extreme’s network AI tools?

Platform ONE is Extreme’s management platform for wired, wireless, fabric and security operations. Extreme says its Agent ONE interface can investigate issues and provide contextual recommendations, meaning the AI is used to manage or assist network operations.

Exchange zones would extend that model beyond network administration. They are intended to run AI applications closer to the data generated by networked operations, using capacity in networking equipment alongside a local compute platform or customer server. Extreme Exchange, the company’s marketplace for AI tools, agents and applications, is expected to supply those applications.

Extreme has also reported early results for Agent ONE Coworker, saying customers saved about 1,000 troubleshooting hours in its first month; 41% of Platform ONE customers used it for daily tasks; and users responded to 98% of critical alerts. Those figures are company-reported and concern the existing network-operations product, rather than Exchange zones.

What remains unknown?

The preview leaves material implementation questions unanswered. Network World reported that Extreme had not specified the qualifying hardware or explained how it would calculate available processing capacity. Extreme also has not yet provided evidence of performance, costs or operational outcomes for Exchange zones.

Those details are necessary for prospective buyers to assess workload fit and the potential effects on network operations. Until the planned 2027 launch, Exchange zones remain a product roadmap centered on using deployed networking equipment as part of a local AI infrastructure layer.

This story draws on original reporting from SiliconANGLE.