Scaling Enterprise AI Inference On-Premises: Extreme Networks and LIQID Inc. Partner to Empower AI Workloads

Extreme Networks and LIQID have announced a new strategic partnership to help enterprises deploy, scale, and manage production‑grade AI inference workloads on‑premises. Together, the companies will enable customers to deploy networks built for the realities of modern inference: always‑on services, variable demand, and strict requirements for consistency, reliability, and control.

This combined approach closes a critical gap in many AI deployments. While inference orchestration platforms can schedule inference workloads, consistent performance at scale requires elastic compute and deterministic networking underneath. The Extreme–LIQID partnership will align compute, network, and orchestration into a cohesive solution purpose‑built for enterprise AI inference.

Why On-premises AI is having its moment

For years, the default playbook for AI infrastructure was centered on public cloud. Data sovereignty requirements, regulatory mandates, and the sensitivity of proprietary training data have kept enterprises away from cloud consumption. Today, as mature, pre‑trained models become widely available, the focus is shifting from experimentation to execution. The opportunity to embed AI‑driven workflows into core enterprise operations—and deliver tangible business outcomes—is now rapidly becoming a practical reality.

As enterprises move from AI experimentation to production, the underlying economics are shifting as well. Token‑based pricing, data egress fees, and the rapid growth of GPU‑intensive workloads driven by newer models are making always‑on cloud AI consumption increasingly unsustainable. At high data volumes, tokenomics simply do not favor sustained cloud usage. At the same time, the latency requirements of many enterprise workflows increasingly rule out remote model consumption, driving the need for on‑premises AI deployment.

AI workloads are elastic

Enterprises rarely start AI initiatives with a clean‑slate infrastructure. AI adoption usually involves integrating AI inference workloads into existing CPU-centric infrastructure. Such infrastructure may need to be segmented or in rare cases re-designed to enable AI inference workloads.

GPU sizing is yet another problem that is hard to scale. Production inference workload GPU consumption is dependent not only on the model size, but also on the throughput required (number of tokens per second). An infrastructure that is based on rigid GPU sockets and networking form factors may run out of capacity at peak times and be idle at other times. Given the high cost of GPUs, usage elasticity needs to be factored in.

A production-grade AI data center requires two core building blocks working in harmony:

• A lossless, high-performance network fabric that supports zero oversubscription, low latency, and east-west bandwidth at scale.
• An elastic, software-defined compute foundation that can compose, scale, and reallocate GPU and memory resources in real time to match shifting workload demands.

This is precisely what the Extreme Networks and LIQID partnership is designed to deliver.

Extreme solution + LIQID Matrix

LIQID Matrix® is the first and only software-defined composable infrastructure platform that offers a single, unified UI for GPU, Memory, and Storage composability. Built for the performance and agility demands of AI, HPC, and modern enterprise applications, LIQID Matrix delivers the speed of the cloud with the control and cost-efficiency of on-premises infrastructure.

Extreme’s end‑to‑end solution—combining Extreme software, Extreme hardware platforms, and Extreme Fabric— brings proven automation and reliability that power data centers and communication service providers to enterprises. These elements provide consistent security, rapid deployment, and operational simplicity enterprises need to build and operate AI workloads.

Through this partnership, LIQID and Extreme will help organizations create an AI infrastructure that has a low deployment risk, is cost-optimal, and is maximally utilized, supporting the rapid deployment and operational simplicity enterprises need to scale AI confidently from pilot initiatives to production environments.

Learn more

This partnership will be showcased in a dedicated session at Extreme Connect 2026, where attendees will experience a hands-on walkthrough of the Extreme and LIQID AI partnership. The session will answer the questions every IT leader is asking: why on-premises AI is winning, where to begin, what to watch for as you scale, and how this partnership will deliver a resilient, cost-efficient, and production-ready AI infrastructure foundation.

Learn more about Extreme Fabric and explore LIQID Matrix.