Equinix, NVIDIA, Together AI target enterprise inference with Q1 2027 launch

Equinix is betting its 280-site interconnection network becomes the default home for enterprise AI inference, pairing NVIDIA compute with Together AI's open-model platform in a distributed service due in the first quarter of 2027.

"Equinix Inference Exchange will enable architectures that are neutral by design, open by default and engineered for exceptional performance," Adaire Fox-Martin, chief executive officer of Equinix, said in a statement.

The three-way split puts Equinix in charge of power, advanced cooling and day-two operations, linked through its Equinix Fabric to clouds, networks and AI providers. NVIDIA anchors the build with its Enterprise Reference Architectures, tuned to maximize AI factory throughput and cut token cost, while Together AI runs the platform across more than 200 open-source models in both multi-tenant and dedicated single-tenant deployments.

The launch extends NVIDIA's compute franchise from training into inference, a market the company says now out-earns training after the two sat roughly even 18 months ago. For Equinix, the service deepens its role as the neutral exchange where eight of the top 10 AI model providers and nine of the top 10 AI cloud providers already sit, giving it a differentiated entry into enterprise open-model inference as the AI data center buildout shifts from construction toward running workloads.

Why Inference Is Moving to the Network Edge

Enterprises are discovering that inference behaves differently from training. Training can run in a handful of hyperscale campuses, but inference must sit close to the users, data and applications it serves, spread across clouds, models and geographies. That creates a latency and compliance problem for regulated industries such as financial services and healthcare, where data residency rules bar shipping workloads to a distant model vendor.

Equinix Inference Exchange is built to answer that, supporting metro-edge inference, open-model migration and sovereign AI. Customers can run inference inside local data centers that satisfy residency requirements while keeping control over where data is processed. The company's footprint — more than 280 data centers across 77 metros, roughly 230 cloud on-ramps and over 10,500 interconnected businesses — gives it the physical reach that a pure software play cannot match.

NVIDIA's Rack-Level Push Completes the Loop

The partnership is one leg of a broader NVIDIA strategy to define the full stack of AI infrastructure rather than just sell chips. The company has agreed to buy open-source platform Hugging Face for $12.9 billion, licensed Groq's LPU technology for its next-generation Vera Rubin architecture, and subscribed to $3.5 billion of MediaTek convertible bonds to co-develop rack-level systems. Together, the moves cover the chain from model training to inference deployment.

For Equinix, the deal arrives as investors weigh its growth against a heavy capital bill. The company carries a market value of about $101 billion with a price-to-sales ratio near 10.3 times, above its historical median, and remains free-cash-flow negative as it funds expansion. Equinix shares have gained 4.8 percent over six months against a 1.8 percent decline for its industry, and Zacks rates the stock a Hold. Peer Digital Realty, which also courts AI tenants, carries a Zacks Buy rating.

Equinix has not disclosed financial targets for Inference Exchange, and the service will not reach customers until the first quarter of 2027. The bet is that by then, the enterprise shift from experimentation to production will have made distributed inference a growth wave Equinix's interconnection network is built to capture.

This article is for informational purposes only and does not constitute investment advice.