The partners are bundling reference AI hardware, networking, and rented data center space into one deployment stack, betting that infrastructure outside AWS, Azure, and Google can win on cost and control for some workloads.
The pitch lands on a single bet: that putting AI hardware in a third-party data center, behind that data center's private network, and running it on a reference architecture from a chipmaker is a better deal for some enterprise AI workloads than renting the same hardware from Amazon, Microsoft, or Google.
On June 16, 2026, Equinix, Cisco, and NVIDIA announced an expanded collaboration to make that pitch concrete. Equinix will host Cisco's "Secure AI Factory with NVIDIA" across its global footprint of colocation data centers, with standardized blueprints and automation the partners say will simplify rollouts. Equinix is also pairing with systems integrator Presidio to build a "Programmable AI Technology Hub (P.A.T.H.) Lab" inside its facilities, where customers can test and validate the stack before committing to a full deployment.
Colocation, the model Equinix sells, is rented floor space, power, cooling, and cross-connects in a carrier-neutral data center. A customer brings its own servers or, increasingly, rents GPU capacity from a partner inside the building. The promise is direct: the same NVIDIA chips, the same Cisco networking, the same private connectivity to other enterprises, with no hyperscaler margin and no hyperscaler lock-in. Cisco and NVIDIA gain a sales channel that does not run through their largest customers' competitors.
The same three companies, however, are deeply embedded with the hyperscalers. Cisco sells networking into every major cloud. NVIDIA's GPUs are the silicon foundation of AWS, Azure, and Google's AI services. The colocation pitch is, in part, a hedge: a way to reach enterprises whose procurement, compliance, or workload economics point them away from public cloud for parts of their AI stack. The partners are not abandoning the hyperscalers. They are widening the funnel.
What the press release does not yet show is whether any enterprise is actually moving production AI workloads off the hyperscalers and onto the Equinix-hosted stack at scale. The P.A.T.H. Lab is a testing environment, not a production deployment. Presidio is positioned as the systems integrator, which means customer adoption runs through a reseller relationship rather than direct enterprise sales by Cisco or NVIDIA. Gordon Mackintosh, Equinix's SVP of global partner sales and ecosystems, framed the bet this way in the release: "The success of enterprise AI starts with its physical foundation."
The economics of colocation for AI depend on power, density, and cooling that older facilities were not designed to provide. The useful test over the next two quarters is whether a named customer runs a named workload on the Equinix stack with public numbers on cost per training hour or per inference. Until that arrives, the partnership is best read as a credible go-to-market experiment from vendors that need more than one distribution channel for enterprise AI.