Hardware · Facility · Network · Security

The physical layer
your models depend on.

A complete technical overview of our compute architecture — from the data center facility through hardware configuration, network fabric, power infrastructure, and physical security.

Facility

Mission-Critical
Pacific Northwest

We operate exclusively from fault-tolerant facilities selected from among the world's leading data center operators. Power, redundancy, and renewable energy certification are non-negotiable selection criteria.

Active Locations
Oregon · Hillsboro / Portland
Phase 1 — Primary Facility
Active
Washington · Seattle
Phase 2 — Geographic Expansion
Planned
Availability 99.982% Availability · Uptime Institute Tier-3 Standard · contractually bound
Maximum 1.6 hours unplanned downtime per year · SLA breach = contractual remedy per MSA
Redundancy 2N fault-tolerant architecture
Dual independent power feeds · Dual cooling systems
No single point of failure — concurrently maintainable (Tier-3 standard)
Power Infrastructure Dedicated power capacity per client
Dedicated PDUs — zero shared distribution
Dual diverse utility feeds from independent grid paths
Connectivity Dual diverse fiber entry points — independent physical routes, independent carriers
Private dark fiber connectivity available for sovereign workloads upon engagement
SLA Enforcement SLA breach triggers contractual remedies per the executed MSA — not service credits
PUE overrun clause with operator penalty
Operator Selection Tier-3 equivalent carrier-neutral colocation operator · North America · institutional-grade
Specific operator identity disclosed to qualified clients under NDA during engagement
Why Pacific Northwest · Hydroelectric-dominant grid (~80%+ renewable baseline)
· Naturally cold climate — reduced DLC load
· Dense, established fiber backbone
· Stable, low-cost power grid
· 36-month fixed electricity pricing contractually locked

"Verde Compute operates exclusively from mission-critical, fault-tolerant data centers in the Pacific Northwest — selected from among the world's leading operators. Facilities are evaluated on redundancy architecture, power infrastructure, SLA accountability, and 100% renewable energy certification. We do not operate in any facility that fails to meet this bar."

Hardware

NVIDIA Blackwell.
Nothing else.

We deploy on NVIDIA Blackwell B200 architecture via NVIDIA Priority Allocation Track — in partnership with a Tier-1 NVIDIA-Certified OEM Partner. No prior generation. No hybrid configurations. Hardware generation commitments are subject to NVIDIA supply allocation and delivery timelines. Final configuration is confirmed at engagement.

GPU NVIDIA B200 — Blackwell Architecture
NVL72 Pod Configuration
Pod Composition 1,152 × B200 GPU per standard deployment
576 × Grace CPU integrated
Custom configuration available upon engagement
GPU Memory ~214 TB HBM3e total addressable
Coherent memory architecture across the full pod
NVLink Fabric Full NVLink Switch interconnect across all GPUs within pod
No PCIe bottleneck — native GPU-to-GPU bandwidth
Validation Minimum 72-hour factory burn-in — independently certified
Certificate of validation provided prior to client go-live
Allocation Each pod is assigned exclusively to a single client
No resource sharing · No workload co-tenancy · No oversubscription
Blackwell Architecture NVL72 Pod HBM3e Memory NVLink Mesh 72hr Burn-in Dedicated Only
Network

Fabric built
for scale.

From internal GPU-to-GPU bandwidth to external client connectivity — every layer of the network is engineered for the demands of frontier AI workloads.

Internal Fabric NVIDIA NVLink — full-mesh GPU interconnect
NVIDIA Quantum-2 InfiniBand — cluster fabric
Sub-100ns latency across the pod
Border Network 400G Arista / Cisco border routing
Per-client isolated network segment — no cross-client traffic
Facility Connectivity Dual diverse fiber entry — two independent physical routes, two independent carriers
Single fiber cut cannot affect operations
Client Connectivity Standard: Encrypted corporate VPN — enterprise-grade tunneling
Sovereign: Dedicated private dark fiber from client premises to Verde facility — available for defense, sovereign AI, and restricted or sensitive workloads. Coordinated through leading network operators. Client-specific.
Isolation Full network isolation per client at hardware and software layers
"Your data never leaves your cage."
Power & Cooling

~2 MW
dedicated capacity.

Direct-to-chip liquid cooling enables sustained full-load operation. Power is provisioned exclusively — no shared distribution, no shared cooling loops.

Cooling Architecture Direct-to-chip liquid cooling (DLC)
Sustained full-load operation — no thermal throttling
Power Capacity ~2 MW dedicated power per standard deployment
Dedicated PDU allocation — zero shared distribution
Power Redundancy 2N dedicated power paths — two fully independent feeds
UPS backup on both paths · Generator on both paths
No single path failure can interrupt operations
PUE Commitment PUE overrun clause — operator financial liability if PUE > 1.15
Contractually enforced, not aspirational
Energy Rate Structure 36-month fixed electricity rate — contractually locked
No variable pass-through. No energy surprise billing.
Energy

100% Renewable.
Certified.

Oregon's hydroelectric-dominant grid provides one of the cleanest commercial power profiles in North America. We certify this commitment, not just claim it.

Renewable Commitment 100% Renewable Energy — contractually committed
Oregon hydroelectric grid (~80%+ renewable baseline) + Green Tariff coverage
Certification I-REC Committed — International Renewable Energy Certificate
Issued and retired annually from first operational year · Accepted by A+ ESG frameworks
Carbon Status Carbon-neutral target — I-REC committed from first operational year
Automated zero-carbon reporting for client ESG documentation
ESG Compatibility Fully compatible with A+ credit-rated enterprise ESG mandates
Scope 2 emissions reporting support included

Verde Compute is pursuing direct investment in renewable energy generation — moving from green energy consumption toward green energy ownership. This is a long-term vision commitment, not a current operational claim.

Physical Security

Air-gapped.
Biometric. Monitored.

Physical security is not a feature — it is a prerequisite. Your hardware lives in a cage that only authorized personnel can access, under continuous surveillance.

Dedicated Air-Gapped Cage

Your hardware occupies a physically isolated, locked cage within the facility. No other client hardware is co-located in your segment. Zero physical sharing.

Biometric 3-Factor Access

All physical access requires three-factor biometric authentication. Access logs are immutable and available for client audit upon request.

7/24 4K Thermal CCTV

Continuous 4K and thermal imaging surveillance across all facility areas. Footage retained per contractual and compliance requirements.

TEMPEST & EMP Shielding

Available for defense and classified deployments. Electromagnetic emanation controls to TEMPEST standards upon client specification.

Data Isolation & Encryption at Rest

All client data is isolated at hardware level and encrypted at rest. Chain of custody documentation provided. NIST SP 800-88 data erasure upon engagement conclusion.

Shared Responsibility Model

What Verde Compute owns.
What you own.

We believe clarity about responsibility is the foundation of trust. Here is exactly what we control and what remains yours.

Verde Compute Responsibility

  • Physical hardware procurement, deployment, and lifecycle
  • Data center facility and SLA management
  • Power, cooling, and network infrastructure
  • Hardware monitoring, alerting, and incident response
  • NVIDIA software stack, driver management, MLOps tooling
  • Engineering team staffing and NVIDIA certification
  • Physical security — access, CCTV, cage integrity
  • Independent audit reports to all stakeholders
  • Renewable energy certification and ESG reporting

Client Responsibility

  • AI models, weights, and training datasets
  • Application-layer software and orchestration logic
  • Data classification and access policy definition
  • Client-side encryption key management
  • Compliance with applicable data residency regulations
  • Client-side network configuration and VPN endpoints
  • Model governance and output quality assurance
  • Contractual financial obligations per Take-or-Pay
Vendor Standards

Every partner.
Same standard.

Every vendor in our supply chain is contractually bound to the same quality and response time standards we commit to our clients. No exceptions.

Hardware OEM Partner 72-hour factory burn-in · independently certified
4-hour on-site service SLA
Consignment spare parts depot on-site
Liquidated damages for delivery delays
NVIDIA Mission Critical Support: 1-hour Severity 1 response
Dedicated Technical Account Manager (TAM)
Resident Systems Engineer on-site
36-month LTS — version freeze for client stability
Custom CUDA firmware & driver validation
Facility Operator 99.982% Uptime SLA (target) · financial compensation for breach
2N dedicated power path + cooling
36-month fixed electricity pricing
PUE overrun financial penalty clause
100% renewable energy ESG certification
Next

Explore the full service layer.

The hardware is one part of the package. See how we wrap it with engineering, software, and governance.

Explore Our Services →