Can “neutral” colocation capture AI inference profit—now that Nvidia is building inside Equinix’s fabric?

On Sept. 2, 2026, Equinix announced Equinix Inference Exchange, positioning it as an enterprise inference program delivered through Equinix’s global infrastructure and interconnection (“Equinix Fabric”). The offer is explicitly framed as neutral by design, open by default, and targeted at accelerating the move from AI experimentation to production inference.

Historically, hyperscalers have monetized AI not just by selling compute, but by bundling the end-to-end path: model-serving frameworks, low-latency networking, and distribution/serving tooling. Equinix’s Sept. 2 announcement tries to replicate the “bundle” outside the hyperscaler boundary—by pairing Nvidia’s reference architecture (compute/system-level optimization) with Together AI’s inference platform (software/runtime-level optimization) and Equinix’s fabric (connectivity + operational placement).

Equinix is a colocation and interconnection platform company; its financial performance depends on how much workload demand turns into billable physical and virtual footprints (capacity + interconnection). The partnership matters for investors only if it reduces friction for enterprises to deploy inference workloads on a neutral platform instead of a hyperscaler-served endpoint.

What’s still unproven from the announcement: the size of bookings/revenue impact, pricing, and whether large inference buyers will standardize on this neutral exchange for production. The partnership is strong on architecture + integration signaling; investors should treat any margin-capture outcome as a measurable-to-prove-by bookings and attach metric, not a near-term certainty.

Equinix’s Nvidia/Together AI bundle is effectively an orchestration play. Nvidia’s role is to anchor system-level reference architectures; Together AI’s role is to provide the inference platform experience; Equinix’s role is to make the physical placement and network path predictable at enterprise scale. The more enterprises treat inference as a production operation with strict performance and security constraints, the more they will pay for integration that reliably hits latency and throughput targets.

The partnership’s core bet is that enterprises want open-model inference and multi-provider connectivity, but they also want production-grade integration speed and performance. If Equinix successfully becomes the default inference deployment path for a meaningful subset of enterprise workloads, it can capture a larger share of the AI data-center boom’s economic value—specifically the interconnection and hosting workflow profits that hyperscalers often keep inside their own cloud inference stacks.

Conversely, if enterprises keep choosing hyperscaler-served inference endpoints for production—using neutral sites mainly for dev/test—then Equinix’s partnership will be credibility-positive but economics-limited. The differentiating KPI is whether production inference workloads route through neutral connectivity at scale.