Nvidia, Equinix, and Together AI partner to deliver enterprise AI inference services.

Inference Exchange will support more than 200 open-source models, enabling enterprises to run a wide range of models on the platform with Together AI operating the inference stack across multiple sites via Equinix Fabric.
Fabric One uses open specifications developed by AWS and Google Cloud for AI connectivity, signaling a standards-based, cloud-agnostic approach to linking AI workloads across Equinix facilities.
Equinix Inference Exchange footprint includes more than 280 data centers across 77 metros and 230 cloud on-ramps, with availability expected to begin in Q1 2027.
The partnership is pitched as giving mid-market firms a neutral, carrier-dense colocation layer to run Nvidia GPUs with Together AI without locking into a single hyperscaler.
Nvidia, Equinix, and Together AI have formed a new partnership to help enterprises run open-source AI models without relying on big cloud providers. Equinix is launching Inference Exchange, a platform that will support more than 200 open-source models across its global data centers. The service marks a shift in how mid-market companies can access cutting-edge AI computing power.
The partnership tackles a major pain point: enterprises want AI inference capabilities without getting locked into Amazon, Google, or Microsoft's cloud ecosystems. TechBuzz reports that Equinix's 280+ data centers across 77 cities give companies a neutral ground to run Nvidia GPUs. Services will begin rolling out in the first quarter of 2027.
Nvidia supplies the GPUs and software stack—the foundation for running AI models. Together AI operates the actual inference platform that processes queries on those GPUs. Equinix provides the real estate: its global network of data centers and connectivity services that tie everything together. This division of labor lets each company focus on what it does best.
The setup avoids the old model where a single cloud giant controls every layer. ITBrief notes that enterprises can now choose dedicated single-tenant deployments or share multitenant resources depending on their security and cost needs. This flexibility appeals to mid-market firms that need more control than hyperscale providers typically offer.
Equinix Inference Exchange will operate across more than 280 data centers in 77 metropolitan areas with 230 cloud on-ramps—direct connections to other cloud platforms. MarketScreener reports the infrastructure uses Fabric One, an open connectivity standard created by AWS and Google Cloud. This standards-based approach means workloads can move between facilities without vendor lock-in.
The distributed footprint is a key advantage over centralized cloud alternatives. Enterprises can run inference closer to where their data lives, reducing latency and bandwidth costs. By Q1 2027, companies in virtually any major metro will have access to Nvidia GPU compute through Equinix's neutral colocation layer.
Amazon, Google, and Microsoft dominate cloud AI services, but they keep prices high and lock customers into their ecosystems. Mid-market enterprises increasingly want alternatives. Nvidia and its partners are betting that a neutral data-center model—where companies choose their own inference software and GPU provider—will win them over.
This partnership signals a broader trend: enterprises no longer accept being forced to use one provider's entire stack. By supporting 200+ open-source models on neutral infrastructure, Equinix and its partners offer the flexibility and cost efficiency that hyperscalers have historically denied. For many mid-market firms, that's a game-changing option.
Publishers
14
Articles
22
Reach
36