Reserved
1, 2, and 3-year reserved contracts with locked-in pricing and capacity.
GPU-as-a-Service
Institutional-grade GPU capacity. Reserved, transparent, sovereign-aware.
Claymont operates rack-scale Blackwell and Hopper deployments inside our own AI data centres — power-secured, liquid-cooled, and underwritten on long-duration offtake. Reserved capacity, on-demand burst, and financed structures for institutional buyers.
Compute fleet
A transparent rate card.
Indicative reserved pricing in USD, ex-egress. Hover any GPU to preview the hardware; on mobile, tap to expand. On-demand, 1-year, and 3-year terms available on request.
- Memory
- 13.5 TB HBM3e (rack)
- Interconnect
- NVLink 5 fabric
- Workload
- Frontier rack-scale training
- Memory
- 13.5 TB HBM3e (rack)
- Interconnect
- NVLink 5 fabric
- Workload
- Rack-scale training
- Memory
- 288 GB HBM3e
- Interconnect
- NVLink 5 / 1.8 TB/s
- Workload
- Frontier training
- Memory
- 192 GB HBM3e
- Interconnect
- NVLink 5 / 1.8 TB/s
- Workload
- Training, fine-tune
- Memory
- 141 GB HBM3e
- Interconnect
- NVLink / 3.2 Tbps IB
- Workload
- Long-context training
- Memory
- 80 GB HBM3
- Interconnect
- NVLink / 3.2 Tbps IB
- Workload
- Training & inference
- Memory
- 48 GB GDDR6
- Interconnect
- PCIe 4
- Workload
- Inference, rendering
- Memory
- 80 GB HBM2e
- Interconnect
- NVLink / 600 GB/s
- Workload
- Inference, mid-tier training
Pricing is indicative for reserved capacity and excludes egress, storage, and managed services. Final terms reflect region, term length, density, and offtake structure.
Reference build
A 36 MW rack-scale block.
- IT load
- 36 MW
- Rack fleet
- 256 × GB200 NVL72
- GPU count
- 18,432 Blackwell GPUs
- Cooling
- Direct-to-chip + rear-door HX
- Power
- N+1, 100% renewable PPA
- Fabric
- 400 GbE east-west, dual-path 800 GbE uplink
Commercial terms
Built around offtake, not spot.
On-demand burst
Elastic capacity above reservation, billed per GPU-hour.
Sovereign suites
Single-tenant deployments compatible with export-controlled and sovereign requirements.
Egress & storage
Transparent line-items for object, block, and high-throughput storage.
GPU financing
GPUs underwritten as long-duration infrastructure.
The 2025–26 AI cycle has moved from cash-funded fleets to GPU-backed capital structures. Claymont arranges and structures financings against contracted compute, alongside operator-grade execution.
Reserved capacity, financed
1, 2, and 3-year reserved contracts institutional buyers can underwrite as offtake.
GPU-backed facilities
Structuring and arrangement of debt against contracted GPU revenue — delayed-draw term loans, ABS-style tranching, IG-eligible structures where the offtaker supports it.
Sale-leaseback
GB200 / GB300 NVL72 racks placed with institutional lessors; Claymont retains operatorship.
Vendor & co-investment
Strategic equity from silicon and hyperscaler partners alongside institutional debt.
Sovereign & export-controlled
Single-tenant structures compatible with US export-licence regimes.
Indicative structures. All financing subject to credit, jurisdiction, and counterparty review.
Why Claymont
Compute owned end-to-end.
Power-first siting
Sites originated against firm power and PPA-backed renewables, not waitlists.
Sovereign-aware deployment
Single-tenant suites and export-licence-aware build-out for regulated buyers.
Operator on-site
We design, build, and operate. One team, one accountability line.