GPU Infrastructure

Engines of abundant intelligence.

Curiosity AI deploys the latest generation of NVIDIA and AMD GPUs across AI Factories for training workloads, inference, frontier AI modelling, generative AI and high-performance computing.

Flagship Training

NVIDIA Blackwell B300 SuperPOD

5MW air-cooled cluster architecture featuring fifth-generation Tensor Cores and 3.2 Tbps Quantum-2 InfiniBand for trillion-parameter frontier training.

Architecture: NVIDIA BlackwellMemory: 288GB HBM3e / GPUInterconnect: 3.2 Tbps NVLink 5 + InfiniBandCooling Model: Air-Cooled Tier-IV Ready

Ultra-Density Inference & Training

NVIDIA GB300 NVL72 Liquid-Cooled

10MW liquid-cooled compute pods with direct-to-chip CDU loops, delivering maximum computational density per square foot with ultra-low PUE.

Architecture: Grace Blackwell GB300Pod Density: 72 GPUs / Liquid-Cooled RackPUE Efficiency: < 1.22 PUECooling Model: Direct-to-Chip Liquid Cooling

Open Sovereign Compute

AMD Instinct MI325X Cluster

High-bandwidth open compute platform engineered for massive memory capacity, large context windows, and cost-effective enterprise inference.

Architecture: AMD CDNA 3Memory: 256GB HBM3e (6 TB/s)Fabric: UltraScale Spectrum-X RoCESoftware Stack: ROCm 6.0 Open Ecosystem

Workloads

Built for every stage of the AI lifecycle.

TrainingInferenceFrontier AI modellingGenerative AIHigh performance computing

Phase 1 configuration

5MW. Four clusters. One deployment.

The opening deployment is four SuperPOD clusters of NVIDIA Blackwell B300 GPUs (2,304 GPUs in total), installed in a partner AI Factory on a five-year colocation model.

SuperPOD clusters in Phase 1

GPUs per cluster

Total GPUs, Phase 1

Nodes per cluster · 8 GPU/node

Reserve long-term GPU capacity.

3–5 year GPU-as-a-Service and bare-metal reservations for neo-clouds and enterprise AI.

For customers Talk to us