How are enterprises balancing cloud and colocation for AI workloads?

Enterprises increasingly split training and inference across environments. Training is often placed in public cloud to use elastic GPU capacity and specialized networking, while inference and data processing move to colocation for predictable costs, data locality, and compliance. This hybrid approach leverages interconnection to major clouds and data sources. Recent updates from AWS, Microsoft, and Google, paired with landlord guidance, indicate this trend is solidifying across large enterprises.

What rack densities are enterprises targeting for new AI deployments?

Many enterprises are planning for 30–60kW per rack in AI-focused expansions, enabled by liquid cooling readiness and upgraded power distribution. Digital Realty and Equinix outline designs for high-density rows and validated cooling strategies. These configurations support GPU clusters and low-latency interconnects for inference pipelines and model serving. The push to higher densities is most visible in late 2025 and early 2026 enterprise plans and colocation procurement.

What role do sovereign cloud controls play in deployment strategies?

Sovereign cloud controls, particularly in the EU, are shaping workload placement decisions to meet residency and regulatory requirements. Google Cloud and Microsoft have enhanced regional controls and compliance features, which enterprises combine with colocation for data gravity and cost. As EU energy and reporting frameworks evolve, enterprises favor facilities offering transparent efficiency metrics and governance, enabling compliant operations without sacrificing performance or time-to-deploy.

Are enterprises adopting liquid cooling widely or selectively?

Adoption is growing but remains selective, typically focused on the densest AI racks where thermal limits are most acute. Industry research suggests liquid cooling pilots span roughly a fifth of AI estates, with direct-to-chip approaches gaining traction. Landlords provide deployment playbooks and modular retrofits to accelerate adoption. OEM ecosystems from NVIDIA, Supermicro, HPE, and Dell support these pilots with liquid-ready designs to streamline integration and operations.

Which vendors are influencing enterprise deployment timelines the most?

Hyperscalers and OEMs are central to deployment speed. AWS, Microsoft, and Google provide elastic GPU capacity and orchestration tooling that guide hybrid strategies. NVIDIA’s platform ecosystem, along with Supermicro, HPE, and Dell, delivers validated, dense, and liquid-ready server configurations. Landlords like Equinix and Digital Realty complement these with power, cooling, and interconnection readiness, compressing time-to-live for AI workloads across regions and industry verticals.

Equinix and Digital Realty Report Enterprise Shift to High-Density Hybrid Deployments

Enterprises accelerate hybrid and colocation strategies to support AI workloads, pushing rack densities to 30–60kW and expanding liquid cooling pilots. New offerings from AWS, Microsoft, and Google shape deployment plans, while analysts flag rising sovereign and edge requirements.

Published: January 11, 2026 By Marcus Rodriguez Category: Data Centers

Equinix and Digital Realty Report Enterprise Shift to High-Density Hybrid Deployments

Executive Summary

Enterprises scale hybrid colocation with 30–60kW racks for AI, according to recent landlord and analyst updates.
AWS, Microsoft, and Google release new high-density AI infrastructure options that reframe enterprise deployment strategies...
Analyst notes indicate a growing share of enterprise workloads favoring colocation and sovereign cloud controls in Europe..
Enterprises pilot liquid cooling at 15–25% of AI estates to manage heat and efficiency, industry sources suggest.

Enterprise Momentum to Hybrid Colocation

Enterprise infrastructure leaders report a decisive pivot to hybrid deployments pairing colocation with public cloud, citing density, cost control, and proximity to cloud on-ramps. Data center landlords highlight AI-ready designs and interconnection strategies as core to enterprise adoption in Q4 2025–early Q1 2026.. Analysts estimate enterprises are allocating larger shares of new compute and storage spending toward facilities optimized for GPUs and specialized accelerators, particularly for inference at scale.

Equinix and Digital Realty emphasize cross connects to major clouds and high-density power availability as differentiators for enterprise workload placement. Equinix’s interconnection data and regional expansion updates point to increased bandwidth utilization among enterprise tenants in late 2025, driven by AI pipelines and data-intensive applications. Digital Realty underscores purpose-built AI data halls and guidance for liquid cooling deployment, reflecting ongoing adoption pilots among large customers.

Deployment Strategies Shaped by Hyperscaler Offerings

Product updates from major clouds in December 2025 are reshaping enterprise plans for where and how AI compute is deployed. AWS re:Invent sessions and announcements stressed new instance families and networking features geared for high-throughput model training and inference, influencing hybrid placement decisions for performance and cost....

Read the full article at AI BUSINESS 2.0 NEWS