Infrastructure Built for
Dense Compute
Thermae Computatio is designed from the ground up for AI training, HPC research, and high-density colocation — with liquid cooling, high-capacity power, and low-latency networking as first-class design requirements, not retrofits.
Liquid-First Thermal Design
Air cooling hits a wall at high rack densities. Thermae Computatio is designed around direct liquid cooling (DLC) as the primary thermal removal path — enabling the rack densities that modern AI and HPC workloads require, while simultaneously making waste heat capture practical.
Direct Liquid Cooling (DLC)
- Planned rear-door heat exchangers and cold-plate options
- Supports GPU-dense AI training configurations
- Higher thermal efficiency than air at rack densities above ~20kW
- Enables waste heat recovery at usable temperatures
Free-Air Economization
- Pacific Northwest climate enables economizer mode for significant portions of the year
- Reduces mechanical cooling load and associated energy cost
- Contributes to lower Power Usage Effectiveness (PUE) targets
- Geography-driven advantage — no additional infrastructure required
Waste Heat Recovery
- Thermal energy captured from liquid cooling loop
- Transferred to adjacent hospitality asset via Heat Purchase Agreement
- Designed for zero waste heat discarded to atmosphere
- Turns cooling cost center into contracted revenue stream
Liquid cooling is not a future upgrade path at Thermae Computatio — it is the baseline design assumption. Tenants requiring air-cooled configurations can be accommodated, but the facility is optimized for liquid-cooled, high-density deployments.
High-Density Power Delivery
AI and HPC workloads are power-hungry by nature. The facility is being designed around onsite low-carbon power generation, including advanced geothermal, with final generation capacity subject to site-specific resource testing, engineering, permitting, and financing. A future utility interconnection may provide backup capacity or expansion options.
Per-Rack Power Capacity
- Designed to support GPU-dense AI training racks
- Scalable power distribution per cabinet
- Planned support for high-draw accelerator configurations (H100, MI300X class)
- Power density targets subject to final engineering and site-specific resource confirmation
Redundancy & Reliability
- N+1 redundancy target across critical power paths
- Planned UPS and generator backup systems
- Dual utility feed design under evaluation
- Redundancy specifications subject to final engineering
Grid Source & Carbon Profile
- Washington State grid dominated by hydroelectric generation
- Among the lowest carbon intensity utility profiles in the US
- Structural low-carbon advantage — not dependent on RECs or offsets
- Supports tenant Scope 2 emissions reporting
Power Usage Effectiveness
- PUE targets improved by liquid cooling efficiency
- Free-air economization reduces mechanical cooling overhead
- Waste heat reuse further improves effective energy utilization
- Target PUE to be confirmed through detailed engineering
High-Throughput Connectivity
AI training and HPC workloads are as network-bound as they are compute-bound. The project will target carrier-diverse fiber access, 400G-ready internal architecture, and low-latency regional connectivity. Available carriers, routes, and peering options will be verified during site diligence.
Optical Networking
- Planned 400G optical fabric for intra-cluster traffic
- Supports high-bandwidth GPU-to-GPU communication patterns
- Low-latency switching architecture under evaluation
- Designed for scale-out AI training topologies
Carrier Diversity
- Multiple carrier access points planned
- Diverse fiber entry paths under evaluation
- Available carriers and routes to be confirmed through site diligence
- Carrier mix to be confirmed through site development
Interconnect Architecture
- Spine-leaf switching topology planned for predictable latency
- Supports east-west traffic patterns typical of distributed AI workloads
- In-band management network separation
- Architecture subject to final engineering design
Facility Design & Security
Physical security, access control, and facility design are planned to meet the expectations of enterprise and regulated-industry tenants.
Physical Security
- Multi-factor physical access control planned
- Perimeter security and CCTV coverage
- Mantrap / airlock entry design under evaluation
- Security operations aligned with enterprise colocation standards
Raised Floor & Cabling
- Structured cabling infrastructure planned
- Overhead and underfloor pathways under evaluation
- Cable management designed for high-density deployments
- Separation of power and data cabling paths
Fire Suppression
- Clean agent suppression system planned for compute areas
- Early warning smoke detection
- Designed to minimize equipment damage risk
- System design subject to final engineering and code compliance
DCIM & Monitoring
- Data Center Infrastructure Management (DCIM) planned
- Per-rack power metering for tenant billing accuracy
- Environmental monitoring (temperature, humidity, airflow)
- Remote hands and eyes service planned
Mechanical Systems
- Cooling distribution units (CDUs) for liquid loop management
- Precision air handling for mixed-density zones
- Redundant mechanical systems on critical paths
- Building management system (BMS) integration planned
Compliance Readiness
- Design aligned with SOC 2 Type II audit readiness
- Physical security, monitoring, operational procedures, and governance will be developed to support a future SOC 2 examination and ISO/IEC 27001 certification
- PCI-DSS physical security requirements considered
- Compliance certifications to be pursued post-commissioning
Planning-stage disclosure: All specifications on this page represent current design targets and planning assumptions. Final specifications are subject to engineering, utility, and regulatory confirmation. Thermae Computatio is in active development; figures marked "Up to", "Target", "Planned", or "~" are forward-looking and not guaranteed.
Ready to Discuss Your Requirements?
Whether you're evaluating colocation options, planning an AI training deployment, or exploring a development partnership — we'd like to hear from you.