AI Compute
Modular high-density compute infrastructure on contracted power — sized for training, inference, HPC, GPU hosting, and dispatchable workloads within one operating platform.
Workload Categories
01 / Training
Sustained high-density GPU clusters for frontier-model training. Designed for long-running jobs with deterministic power profiles and consistent thermal envelopes.
Capacity provisioned against contracted utility load — no risk of throttling or de-rating during sustained training runs.
Up to 80 kW/rack · Liquid-cooled options
02 / Inference
Production inference workloads at scale, optimized for cost-per-token and request-level autoscaling across diurnal demand curves.
Co-located storage and network architecture sized for low-latency model serving across enterprise and consumer applications.
20–40 kW/rack · Air-cooled standard
03 / Enterprise
Managed dedicated capacity for enterprise AI customers requiring long-term GPU access without operating their own facilities.
Tenant isolation, contracted SLAs, and enterprise-grade physical and network security from Phase I onward.
Enterprise SLAs structured to tenant requirements.
04 / HPC
Scientific computing, simulation, and modeling workloads with non-uniform power and cooling profiles.
High-density blocks engineered for bursty compute patterns characteristic of CFD, genomics, materials science, and financial modeling.
30–60 kW/rack · Hybrid cooling
05 / Edge AI
Distributed inference and low-latency deployment optionality where the campus serves as a regional aggregation point.
Integrated network connectivity supports tenant edge architectures without requiring separate facility build-outs.
06 / Flexible Load
Bitcoin balancing and grid-responsive compute that absorbs available campus capacity at the margin and curtails on demand.
Generates revenue against unused capacity while providing utilities a dispatchable load resource for grid balancing.
Curtailable · Grid-responsive
Density & Thermal
Compute pods are engineered for the full range of AI workload density — from air-cooled inference racks to liquid-cooled training clusters running sustained at peak load.
Cooling architecture is selected per pod against tenant requirements rather than imposed as a single facility standard, giving operators flexibility to size deployments to actual workload profile.
Target Tenants
01
GPU cloud providers and inference platforms requiring large contracted blocks of high-density capacity with predictable cost structures.
02
Enterprise AI teams, research institutions, and HPC operators seeking dedicated managed capacity without operating their own facilities.
03
Strategic overflow capacity for hyperscalers and grid-responsive bitcoin operators sharing the same infrastructure under dispatchable load agreements.
Tenancy
Tenant inquiries are routed directly to the platform team. Capacity availability, workload fit, and indicative commercial terms are discussed under NDA.