Together AI enters new distribution deal with Equinix and NVIDIA for enterprise inference infrastructure.
The AMW Read
Extends Together AI's post-Series-C distribution push (IBM Cloud deal, now Equinix/NVIDIA) into neutral global colocation infrastructure, a segment-level distribution update rather than a new capability or debate-resolving event.
Together AI enters new distribution deal with Equinix and NVIDIA for enterprise inference infrastructure.
Equinix launched Equinix Inference Exchange, a distributed AI inference program combining NVIDIA's Enterprise Reference Architectures with Together AI's inference platform, which supports over 200 open-source models. The offering runs across Equinix's global IBX data centers — more than 280 facilities in 77 metro areas — connected via Equinix Fabric, and is structured in three layers: Equinix's power, cooling and interconnection infrastructure; NVIDIA's optimized compute architectures; and Together AI's platform layer, which offers both multi-tenant and single-tenant deployment. Equinix also announced a companion product, Equinix Fabric One, for connecting distributed AI environments. Both were unveiled at Equinix's first customer and partner event, Equinix Horizon, in Redwood City, California. Enterprise availability is targeted for the first quarter of 2027.
The deal gives Together AI, an inference platform built around open-weight models, a new distribution channel inside a neutral interconnection fabric that Equinix says already hosts eight of the top ten AI model providers and nine of the top ten AI cloud providers. It follows a run of infrastructure-distribution moves for the company: an $800 million Series C at an $8.3 billion valuation in July and a $240 million multiyear deal with IBM to build a dedicated inference cluster on IBM Cloud in August. Per the AI Market Watch index, which tracks the company under AI Infrastructure with $533.5M in total funding, news mentions of Together AI in our pipeline rose to three in the last 90 days from one in the prior 90 — a pipeline-coverage read, not a market census, but directionally consistent with a company converting a large raise into distribution partnerships rather than product launches alone.
For enterprises, the pitch is metro-edge low-latency inference, a migration path off closed proprietary models, and data-residency controls for regulated or sovereign workloads, without locking into one provider. For builders and investors, it signals that open-model neocloud platforms are increasingly reaching enterprises through neutral colocation and interconnection infrastructure rather than by building out their own global data-center footprint — though the Q1 2027 launch means the actual enterprise reach is not yet tested.
