Sovereign & hybrid AI infra is accelerating
Multiple vendors are shipping large, sovereign and hybrid AI infrastructure — Global AI deployed the largest NVIDIA GB300 cluster in NY, Microsoft says it installed the Vera Rubin NVL72 first in cloud, and Equinix launched a Distributed AI Hub to connect clouds, GPUs and the edge reported reported reported. The pattern is clear: expect more multi‑region, on‑prem and connected‑edge options that platform teams must model for compliance and latency tradeoffs.
Global AI said it completed a GB300 NVL72 rollout at its Endicott, New York site and is “actively deploying 7,000 GB300s,” with an initial 16 MW IT load and a roadmap to hit 100 MW in 2026, 250 MW in 2027 and 1 GW by 2029. (prnewswire.com) Microsoft’s CEO posted that Azure is the first cloud to bring up and validate an NVIDIA Vera Rubin NVL72 system, a rack-scale configuration that vendors report can deliver up to 3.6 exaflops with 72 Rubin GPUs and dense NVLink fabrics. (convergedigest.com) Equinix launched a vendor‑neutral Distributed AI Hub powered by Equinix Fabric Intelligence that exposes private, low‑latency connections across its 280 data centers and integrates with Palo Alto Networks for real‑time threat detection of AI workloads. (newsroom.equinix.com) Equinix cited IDC’s projection that 80% of enterprises will deploy distributed edge infrastructure by 2027 and framed the Hub as a way to unify training, inference and data pipelines that currently span clouds, private data centers and edge sites. (newsroom.equinix.com) Microsoft and infrastructure press coverage note the Rubin bring‑up required multi‑year redesigns for power, liquid cooling and rack‑scale networking, and Global AI’s release says its Endicott deployment integrates compute, networking, storage and liquid cooling for sustained high utilization. (redmondmag.com) Equinix’s neutral interconnection layer and Palo Alto tie‑ins create a pathway to run inference close to users and data while preserving private data flows, and industry writeups flag that Rubin/NVL72 density will force new telemetry and failure‑mode tracking for NVLink, GPU hot‑spots and liquid‑cooling events. (newsroom.equinix.com)
