About NVIDIA L40
The NVIDIA L40 rental price today is $0.95 per GPU-hour, unchanged since yesterday, down 2.8% over the past week and down 2.4% over the past month. At that rate one L40 costs about $690 for a month of 730 hours. Over the past year it has ranged from $0.90 to $1.01. The index is the price of one benchmark contract: L40 48GB PCIe, on-demand, 8-GPU node, US hardware, CPU-neutral. It is standardized across 5 cloud providers and marketplaces, from Vast.ai at $0.57 to CoreWeave at $1.25, and is updated every minute.
The L40 is NVIDIA's Ada Lovelace data-center GPU for graphics and AI, announced on September 20, 2022 to power NVIDIA's second-generation OVX servers for 3D virtual worlds and digital twins. It is a PCIe card with 48 GB of GDDR6 memory, four DisplayPort outputs and no NVLink. It is listed far less widely than the L40S that NVIDIA announced in August 2023: on October 30, 2026 GetDeploying tracked 13 L40 providers, against 40 for the L40S.
Specifications
| Spec | L40 |
|---|
| Architecture | Ada Lovelace, 18,176 CUDA cores, 142 RT cores, 568 Tensor cores |
| Memory | 48 GB GDDR6 with ECC, 864 GB/s |
| FP8 tensor | 362 TFLOPS dense, 724 with sparsity |
| BF16 / FP16 tensor | 181 TFLOPS dense, 362 with sparsity |
| FP32 | 90.5 TFLOPS |
| Interface | PCIe Gen4 x16, 64 GB/s; no NVLink |
| Power | 300 W, passively cooled, dual slot |
| Media, display and virtualization | 3 video encoders and 3 decoders with AV1; 4 DisplayPort 1.4a outputs; virtual GPU support; no multi-instance GPU |
| Typical server | 8 cards per node, as in NVIDIA's second-generation OVX servers |
Source: NVIDIA L40 and its datasheet of January 2023, accessed October 30, 2026; server size from NVIDIA's announcement.
Where the L40 is rented
NVIDIA said on March 21, 2023 that the L40 was available from system builders including Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro. CoreWeave's price list showed an 8-GPU L40 machine in North America at $10.00 an hour on demand, $1.25 per GPU-hour, on October 30, 2026. GetDeploying tracked 13 L40 providers that day, among them CoreWeave, Hyperstack, RunPod, Massed Compute and Oblivus, and its on-demand median covered 12; none was AWS, Azure, Google Cloud or Oracle. Most of its 74 listings were for 1 to 4 GPUs, and 6 providers listed 8-GPU machines.
What this price measures
The index is the posted on-demand price of one benchmark contract, an L40 48GB in an 8-GPU node of US hardware, rented by the hour from GPU clouds and marketplaces. Prices for other node sizes are converted to that contract with the index's node-size factors, and prices from other regions are counted without a regional adjustment for now. Reserved and spot prices are not counted.
On October 30, 2026, GetDeploying listed 8-GPU L40 machines on demand at $0.99 per GPU-hour at Thunder Compute, $1.00 at Hyperstack, $1.05 at Oblivus, $1.10 and $1.51 at Sesterce and $1.25 at CoreWeave; only the Oblivus machine was in stock, and CoreWeave did not report availability. The lowest in-stock price was $0.33, for 2- and 4-GPU machines at Lium. Its on-demand median was $0.84, and it showed no reserved or spot median.
What moves L40 rental prices
- Inference that fits on one card. One L40 can run models up to about 68 billion parameters at 4-bit precision, or 17 billion at 16-bit, with a 32K context, GetDeploying estimates. It has no NVLink, so its cards exchange data over PCIe.
- Graphics and video as well as AI. NVIDIA's datasheet lists rendering, virtual workstations, Omniverse, AI training and video streaming among its workloads, and in March 2023 NVIDIA presented it for image generation, with 7 times the previous generation's Stable Diffusion inference performance.
- The L40S. On GetDeploying's October 30, 2026 medians an L40 rented for about 54% of an L40S's hourly price, $0.84 against $1.57. NVIDIA announced the L40S on August 8, 2023 for AI training and inference as well as graphics.
- Capacity. Where a source shows capacity, this index counts a provider only on days when one of its offers had capacity; its posted prices then all count, so an offer going out of stock does not read as a price change (methodology).
- Export controls. NVIDIA said on October 17, 2023 that a new US rule would require a license for exports of the L40 to China and some other countries.
Price history
- October 2025 to October 2026: GetDeploying's weekly on-demand provider median was $1.075 per GPU-hour, from 8 providers, in the week of October 27, 2025, when its public series starts. It peaked at $1.089 in the week of May 18, 2026, was lowest at $0.97 in the weeks of July 13 to August 3, 2026, and was $1.055, from 11 providers, in the week of October 19, 2026, the latest full week, about 2% below the first.
- October 30, 2026: the median posted on-demand price on GetDeploying's L40 page, across 12 providers and a different measure from its weekly series, was $0.84 per GPU-hour, about 7% higher than 90 days earlier and about the same as a year earlier.
This index's history before October 30, 2026, its first day, is GetDeploying's weekly L40 series, rescaled to join the daily index, counting only weeks with at least 5 providers; every week had at least 8.
Futures and benchmarks
No futures contract or Bloomberg ticker covers the L40. CME's filed contracts and the contracts ICE says it will list, with no date set yet, cover the H100 and B200. The GPU futures ICE announced in May 2026 may cover the H100, H200, B200 and RTX 5090 and other GPUs "as the market develops", and the contracts Nodal Exchange plans are for "Hopper, Blackwell and beyond".
FAQ
How does it differ from the L40S? Both have 48 GB of GDDR6 at 864 GB/s and 18,176 CUDA cores, but the L40S has twice the L40's dense tensor throughput at 16-bit and 8-bit precision and draws up to 350 W against 300 W, according to NVIDIA's L40S page and L40 datasheet.
How does it compare with the A100 or H100 on price? On GetDeploying's October 30, 2026 medians it rented for about 44% of an A100's hourly price, $0.84 against $1.90, and 24% of an H100's, $3.54. It has 48 GB of memory against 80 GB, 864 GB/s of bandwidth against 2,039 GB/s for the A100 80GB SXM and 3.35 TB/s for the H100 SXM, about 58% and 18% of their dense BF16 compute, and no NVLink, according to NVIDIA's A100 and H100 pages.
How often is this index updated? It is recalculated every minute, with one value per day in US Central time that is final when the day ends.
Sources
- NVIDIA: L40 product page and specifications
- NVIDIA: L40 datasheet, January 2023
- NVIDIA: second-generation OVX with the L40 announced, September 20, 2022
- NVIDIA: inference platforms and L40 availability, March 21, 2023
- GetDeploying: NVIDIA L40 prices
- GetDeploying: GPU price dataset, weekly price history, CC BY 4.0
- CoreWeave: GPU cloud pricing
- NVIDIA: export licensing, Form 8-K, October 17, 2023
Text reviewed Oct 30, 2026; the figures in it are as of that date. The figures at the top of this page update every minute.
Cite as: Compute Screener, “NVIDIA L40 rental price index”, computescreener.com/gpu/l40. The figures are free to reuse under CC BY 4.0 with credit to Compute Screener.
