Equinix, NVIDIA and Together AI launch Inference Exchange for production AI

What Inference Exchange combines

Equinix says where inference runs now determines its performance, cost and governance. As enterprise AI moves from pilot to production, the company argues, inference has to sit close to the users, data and applications it serves, across every cloud, model and provider in an environment. Choosing where it runs and connecting it to everything workloads depend on adds operational complexity, even as organizations need more control and visibility.

Equinix Inference Exchange is designed for that decision. It combines three elements:

• NVIDIA Enterprise Reference Architectures
• Together AI's inference platform, with more than 200 open-source models
• Equinix data centers connected through Equinix Fabric to the clouds, networks and AI providers that inference relies on

According to Equinix, the result is a faster path from experimentation to production, with secure, low-latency connectivity to the ecosystem customers already use.

Who provides what

Equinix supplies the infrastructure: AI-ready data centers with power, advanced cooling and day-two operations, connected through Equinix Fabric. NVIDIA anchors the build with Enterprise Reference Architectures and accelerated computing that Equinix says is designed to maximize AI factory throughput and lower token cost.

Together AI runs the inference platform on top. It offers multitenant deployments for shared efficiency, as well as dedicated single-tenant environments for workloads that need reserved capacity. Its OpenAI-compatible API means teams can switch models without rewriting integrations, according to Equinix.

Equinix also cites its footprint: 280+ data centers, 77 metros, 230 cloud on-ramps and 10,500+ interconnected businesses. The company says eight of the top 10 AI model providers and nine of the top 10 AI clouds are deployed with Equinix.

Three intended use cases

Equinix outlines three ways enterprises will use the service.

• Metro edge inference: run inference close to users and data for lower-latency AI experiences.
• Open model migration: move production workloads from closed, proprietary models to open-source alternatives. Together AI's platform is reachable over the same Fabric connections used for other providers, so teams can control cost and avoid lock-in.
• Sovereign AI: run AI workloads in locations that meet data residency and sovereignty requirements, with control over where data and inference are processed.

Availability and what it means

Equinix will show the offering at NVIDIA GTC Berlin, Hall 7 booth 7037, on October 20-22. The service itself is due to be available starting in Q1 2027. The source does not give pricing.

For enterprises, the pitch is a single place to run inference near their data while keeping access to multiple clouds, networks and model providers. Adaire Fox-Martin, CEO and President of Equinix, said: "Equinix Inference Exchange will enable architectures that are neutral by design, open by default and engineered for exceptional performance."