GPU compute, rented on terms that match your workload
Burst capacity for training runs and reserved allocations for steady inference, delivered from partner data centres with transparent metering and written quotations.
Request a Quotation
What the service covers
Compute is only useful with the network, storage, isolation and reporting around it. All of these are quoted together.
Burst Training Capacity
Dense multi-GPU configurations with high-bandwidth interconnect, rented for the duration of a training run and released afterwards.
Reserved Monthly Allocation
Predictable capacity for steady inference traffic, reserved on monthly or annual terms with agreed availability.
Network and Storage
Bandwidth, object storage and snapshot options quoted alongside compute, so total cost is visible before commitment.
Isolation and Access Control
Tenant isolation, credential management and access logging configured per engagement and documented in the service agreement.
Usage Reporting
Consumption reported by resource and time period, with alert thresholds so that spend stays within the agreed budget.
Efficiency Data
Energy performance data from partner facilities is made available to customers supporting their own ESG reporting.
Matched to how compute is actually consumed
Training and inference behave very differently. Contracting should reflect that difference rather than forcing both into one shape.
Model Training and Fine-Tuning
Short, intensive runs requiring dense GPU capacity for days or weeks, then released.
Production Inference
Steady, latency-sensitive traffic serving end users, reserved on monthly capacity.
Research and Experimentation
Flexible capacity for universities, labs and product teams validating new approaches.
Rendering and Simulation
Parallel compute for visualisation, engineering simulation and batch data processing.
Efficiency is reported, not assumed
Customers receive energy performance data from the facilities serving their workloads, which can support internal ESG and carbon accounting.
High-Efficiency Fleet
Hardware is selected on performance per watt, and low-efficiency assets are retired on a planned refresh cycle.
Cooling and Scheduling
Containment and liquid cooling are evaluated against measured PUE; scheduling shifts batch work into off-peak windows.
Reported Targets
Efficiency, renewable share and carbon intensity targets are stated as planning targets and reviewed annually.
Common questions
Model and quantity of GPUs, memory and storage requirements, expected rental period, network bandwidth, whether a managed environment is needed, and your target region. With these details we can issue a written configuration and pricing proposal.
Monthly, quarterly and annual terms are available, and short-term project rentals can be arranged subject to availability. The applicable period, pricing and service levels are confirmed in the written quotation and contract.
Usage is metered per application, with input and output tokens counted separately. Monthly quotas, rate limits and alert thresholds can be configured so that spend stays within the agreed budget. Unit pricing is confirmed in the quotation.
Storage media are cleared to the standard agreed in the contract before the asset leaves your control, and a sanitisation record is provided. Where required, media can be retained by the customer under the contract terms.
Tell us your workload
Share GPU model and quantity, memory and storage needs, expected rental period, bandwidth and target region. We will respond with a written configuration and pricing proposal.