Compare your options
Should I buy or rent AI compute?
There is no single right answer. Utilisation, control, data location and how often you expect to upgrade all change it. This tool sets the options out side by side.
Rental cost over the period
Demonstration data
Hourly dedicated
£22,560
2 × NVIDIA H100 80 GB · 400 h/month · 12 months
Monthly dedicated
£30,960
Whole-GPU reservation, idle time included
At roughly 55% utilisation, a monthly reservation is more expensive than paying by the hour for this pattern. Ownership becomes worth modelling once utilisation stays high across a year or more.
Rates shown are demonstration figures held in the site catalogue, not live market pricing. Purchase, refurbished and finance costs are quoted individually — request current pricing before making a commercial decision.
The six commercial routes
| Option | Where it wins | What to watch |
|---|---|---|
| Buy new | Highest control, best economics at high utilisation, retains resale value. | Capital outlay, lead times, and you own the deployment. |
| Buy refurbished | Ownership at a lower entry price where a previous generation suits the workload. | Availability varies, and testing status must be checked per unit. |
| Finance or lease | Spreads cost across the useful life rather than one capital hit. | Subject to status and approval; total cost is higher than outright purchase. |
| Hourly compute | No commitment, ideal for evaluation and short runs. | Most expensive per hour; costs escalate quickly with sustained use. |
| Monthly dedicated | Predictable monthly cost with the whole server reserved. | You pay for idle time as well as working time. |
| Reserved capacity | Guarantees access to specific GPUs for a defined window. | Requires commitment in advance and supplier confirmation. |