Pillar Page
GPU data centers are purpose-built facilities engineered for the power, cooling, and density requirements of AI accelerators. Constellation develops, finances, owns, and operates hyperscale-ready GPU data centers with direct-liquid-cooling, 80 kW rack density, and 25 MVA grid connections — the physical foundation of the AI compute economy.
Traditional data centers were designed for general-purpose IT workloads with rack densities of 8–15 kW and air cooling. GPU data centers are fundamentally different — they require rack densities of 80 kW or more, direct-liquid-cooling infrastructure, and power connections measured in megavolt-amperes rather than kilowatts. These requirements make GPU data centers a distinct infrastructure asset class.
Constellation's GPU data centers are designed from the ground up for AI compute, with power-first site selection in the GCC (leveraging low-cost energy), direct-liquid-cooling infrastructure, and Tier III design standards. Each facility is built to support continuous, high-utilisation AI training and inference workloads at production scale.
25 MVA grid connections and power-first site selection leveraging the GCC's low-cost energy advantage (~USD 0.05–0.06/kWh).
Direct-liquid-cooling (DLC) infrastructure enabling rack densities up to 80 kW — essential for current-generation GPU systems.
current-generation 8-GPU platform platforms in standardised 8-GPU nodes through authorised distribution.
Tier III design standard, 24/7 NOC monitoring, and enterprise-grade physical and cyber security frameworks.
Phased deployment from 3,200 GPUs (Phase 1) to 7,840+ GPUs (Phase 2–3) with modular facility architecture.
Carrier-neutral network access with international fiber-optic routing and subsea cable infrastructure.
Frontier AI model training
Enterprise AI inference
Sovereign AI programmes
Neocloud GPU hosting
Government compute
Research & HPC
Server Platform
Supermicro SYS-822GS-NB3RT
GPU Platform
current-generation 8-GPU platform
Max Rack Density
80 kW (liquid-cooled)
Grid Connection
25 MVA
Design Standard
Tier III
Target SLA
99.99% availability
View our portfolio of GPU data center deployments and investment opportunities.
Explore Projects