Equinix Inference Exchange

Equinix Inference Exchange is an AI inference service built by Equinix — the company that rents out data center space — together with NVIDIA and Together AI. Inference is the process where an already-trained AI model takes a new question or input and produces an answer.

Think of it this way: Equinix is the landlord, and Together AI is the shop owner who greets customers inside that building. When a customer walks into the Together AI "store" and picks one of over 200 open-source AI models, the bill comes under Together AI's name, but the room where the actual computation happens sits inside an Equinix building. Rather than devoting most of its space to a handful of big AI labs the way hyperscalers do, Equinix's strength lies in letting customers of all sizes choose from a variety of compute chips.

The service will mostly run on Equinix facilities built around NVIDIA's B300 Blackwell Ultra chips, with some liquid-cooled facilities also offering the newer Vera Rubin chips. Equinix Fabric One, which simplifies connections across multiple clouds and models, was announced alongside it, and the official launch is set for Q1 2027.

As reported: "The newly announced service is called Equinix Inference Exchange, set to officially launch in Q1 2027." A common misconception is that Equinix itself sells AI models or bills customers directly. Billing and sales are handled by Together AI, while Equinix provides the building and GPUs where the computation actually runs.