Cost to rent L40S for 730 hours (1 GPU)
Compute cost of 730 GPU-hours on one L40S, from each vendor's published on-demand rate, among tracked providers. Compute subtotal only, excludes storage, egress, vCPU, tax. Among tracked providers. Not an availability claim, except rows marked out of stock when checked. Assumes 730 h on 1 GPU.
Lowest on-demand L40S on this page (not marked out of stock): CloudRift L40S 48GB at $6.30 / $63.00 / $459.90 (10 / 100 / 730 h) - variants differ, see tables.
L40S 48GB
Cheapest L40S 48GB among tracked providers: CloudRift at $459.90. Lowest $459.90, median $1,138.80, highest $1,387.73 across 6 rates from 6 vendors (spread $927.83).
| Vendor | Tier | $/GPU-hr | 730 h subtotal | Verified | Source | Link |
|---|---|---|---|---|---|---|
| CloudRift | On-demand | $0.63 | $459.90 | 2026-10-30 13:34 ET | vendor page | Direct link |
| RunPod | On-demand (Secure Cloud) | $1.09 | $795.70 | 2026-10-30 13:36 ET | vendor page | Referral link |
| Nebius | On-demand (from, Intel CPU) | $1.55 | $1,131.50 | 2026-10-30 13:35 ET | vendor page | Direct link |
| DigitalOcean GPU Droplets | On-demand | $1.57 | $1,146.10 | 2026-10-30 13:34 ET | vendor page | Direct link |
| OVHcloud (US) (excl. tax/VAT) | Public Cloud GPU instance, hourly ("Price ( excl. taxes / hour)") | $1.8 | $1,314.00 | 2026-10-30 13:35 ET | vendor page | Direct link |
| Verda (formerly DataCrunch) | On-demand instance (1x GPU) | $1.901 | $1,387.73 | 2026-10-30 13:36 ET | vendor page | Direct link |
What's not included
Storage, egress, vCPU/RAM add-ons, region and tax. Availability is an automated check at the time shown: rows marked out of stock when checked are listed last and left out of lowest, median and cheapest; other rows are not an availability claim. Multi-GPU and 8-GPU-only offers are left out: multi-GPU quotes need whole-instance pricing, coming later. Spot, community, marketplace, serverless, reserved and held or thin prices are never shown here.
Other durations: 10 h · 100 h. Full price list: L40S rental prices. For your own hours or throughput, use the MCP tool estimate_job_cost.