The cheapest L40S on the table right now is $0.200 per GPU-hour at Vast.ai (spot), out of 59 offers from 25 providers. The L40S carries 48 GB of VRAM on NVIDIA's Ada generation. That is peer/marketplace capacity — rented from other users' machines, and not directly comparable to a dedicated instance.
On-demand pricing for the same card spans $0.380 to $3.77 per GPU-hour — a 9.9× spread between the cheapest and the most expensive provider for identical silicon. That gap is the entire reason this table exists.
Cheapest by pricing model: on-demand $0.380 at Lium.io · spot $0.200 at Vast.ai · reserved $0.655 at LeaderGPU.
| Provider | Tier | $/GPU-hr | Pricing | Node | Commit | In stock | |
|---|---|---|---|---|---|---|---|
| Vast.ai | marketplace | $0.200 | spot | 4, 8× GPU | — | yes | rent |
| Lium.io | marketplace | $0.380 | on-demand | 1, 2× GPU | — | yes | rent |
| AWS | dedicated | $0.423 | spot | 4× GPU | — | — | rent |
| AWS | dedicated | $0.452 | spot | 8× GPU | — | — | rent |
| Vast.ai | marketplace | $0.467 | spot | 2× GPU | — | yes | rent |
| Vast.ai | marketplace | $0.467 | on-demand | 4, 8× GPU | — | yes | rent |
| Vast.ai | marketplace | $0.533 | spot | 1× GPU | — | yes | rent |
| Vast.ai | marketplace | $0.534 | on-demand | 2× GPU | — | yes | rent |
| Novita | dedicated | $0.550 | on-demand | 1× GPU | — | — | rent |
| AWS | dedicated | $0.550 | spot | 1× GPU | — | — | rent |
| GPU.ai | dedicated | $0.610 | on-demand | 1× GPU | — | yes | rent |
| LeaderGPU | dedicated | $0.655 | reserved | 8× GPU | 1 mo | — | rent |
| Verda | dedicated | $0.685 | spot | 1–8× GPU | — | — | rent |
| Runpod Community | marketplace | $0.790 | on-demand | 1× GPU | — | yes | rent |
| Runpod Community | marketplace | $0.790 | spot | 1× GPU | — | yes | rent |
| Vast.ai | marketplace | $0.801 | on-demand | 1× GPU | — | yes | rent |
| Prime Intellect | dedicated | $0.820 | on-demand | 1× GPU | — | yes | rent |
| Cudo Compute | dedicated | $0.870 | on-demand | 1× GPU | — | yes | rent |
| Massed Compute | dedicated | $0.880 | on-demand | 1–8× GPU | — | — | rent |
| Spheron | dedicated | $0.960 | on-demand | 1× GPU | — | — | rent |
| Seeweb | dedicated | $0.990 | on-demand | 1× GPU | — | — | rent |
| Runpod | dedicated | $0.990 | on-demand | 1× GPU | — | yes | rent |
| Runpod | dedicated | $0.990 | spot | 1× GPU | — | yes | rent |
| Koyeb | dedicated | $1.20 | on-demand | 1× GPU | — | — | rent |
| Civo | dedicated | $1.29 | on-demand | 1–8× GPU | — | — | rent |
| Verda | dedicated | $1.37 | on-demand | 1–8× GPU | — | — | rent |
| Crusoe | dedicated | $1.50 | on-demand | 1× GPU | — | — | rent |
| DigitalOcean | dedicated | $1.57 | on-demand | 1× GPU | — | — | rent |
| Scaleway | dedicated | $1.71 | on-demand | 1–8× GPU | — | — | rent |
| Beam | dedicated | $1.75 | on-demand | 1× GPU | — | yes | rent |
| OVHcloud | dedicated | $1.80 | on-demand | 1× GPU | — | — | rent |
| AWS | dedicated | $1.86 | on-demand | 1× GPU | — | — | rent |
| Modal | dedicated | $1.95 | on-demand | 1× GPU | — | yes | rent |
| CoreWeave | dedicated | $2.25 | on-demand | 8× GPU | — | — | rent |
| AWS | dedicated | $2.62 | on-demand | 4× GPU | — | — | rent |
| Oracle Cloud | dedicated | $3.50 | on-demand | 4× GPU | — | — | rent |
| Replicate | dedicated | $3.51 | on-demand | 1, 2× GPU | — | yes | rent |
| Replicate | dedicated | $3.51 | reserved | 4, 8× GPU | — | yes | rent |
| AWS | dedicated | $3.77 | on-demand | 8× GPU | — | — | rent |
How much does an L40S cost per hour?
As of the latest scrape, $0.200 per GPU-hour at Vast.ai is the cheapest L40S offer across 25 tracked cloud providers. The cheapest on-demand rate is $0.380 per GPU-hour at Lium.io, and the most expensive on-demand rate is $3.77 at AWS.
Which cloud is cheapest for the L40S?
Vast.ai at $0.200 per GPU-hour on spot capacity. Prices move constantly, so this page is regenerated from the live scrape every 15 minutes. Note that marketplace tiers are peer hardware and are not directly comparable to dedicated capacity.
Is spot L40S capacity cheaper?
Yes — the cheapest spot L40S is $0.200 per GPU-hour at Vast.ai, -47% versus the cheapest on-demand rate of $0.380. Spot instances can be reclaimed by the provider at any time, so they suit checkpointed training and batch inference rather than long-lived services.
How many GB of VRAM does the L40S have?
48 GB, on the Ada architecture. Multi-GPU instances multiply that: an 8× L40S node exposes 384 GB of GPU memory in total.