28 American providers pricing 32 GPUs. 220 listings on machines physically in United States, across 26 GPUs. The cheapest local H100 is $1.99/hr at io.net.
Prices collected 2026-09-18 · every figure is a provider's own published rate
| GPU | Cheapest at | $ / GPU / hr | $ / month | Local median | Global | vs global |
|---|---|---|---|---|---|---|
| $8.60 | $6,192 | $8.601 provider | $8.60 | +0% | ||
| $4.50 | $3,240 | $7.109 providers | $7.85 | −43% | ||
| $4.00 | $2,880 | $6.2517 providers | $6.59 | −39% | ||
| $2.00 | $1,440 | $4.5019 providers | $4.45 | −55% | ||
| $0.50 | $360 | $3.764 providers | $3.72 | −87% | ||
| $1.99 | $1,433 | $3.9726 providers | $3.50 | −43% | ||
| $2.59 | $1,865 | $3.202 providers | $3.20 | −19% | ||
| $2.59 | $1,865 | $3.015 providers | $3.03 | −15% | ||
| $1.67 | $1,203 | $2.293 providers | $3.02 | −45% | ||
| $1.59 | $1,145 | $2.887 providers | $2.88 | −45% | ||
| $1.57 | $1,130 | $2.753 providers | $2.66 | −41% | ||
| $0.50 | $360 | $2.496 providers | $2.59 | −81% | ||
| $1.08 | $775 | $2.2411 providers | $2.05 | −47% | ||
| $1.01 | $724 | $1.833 providers | $1.83 | −45% | ||
| $0.43 | $308 | $2.1521 providers | $1.79 | −76% | ||
| $0.79 | $569 | $2.109 providers | $1.48 | −47% | ||
| $0.69 | $500 | $1.9517 providers | $1.47 | −53% | ||
| $1.10 | $793 | $1.295 providers | $1.29 | −15% | ||
| $0.66 | $475 | $0.827 providers | $0.86 | −23% | ||
| $0.63 | $453 | $0.826 providers | $0.82 | −23% | ||
| $0.32 | $231 | $0.8010 providers | $0.80 | −60% | ||
| $0.12 | $87 | $1.558 providers | $0.79 | −85% | ||
| $0.48 | $347 | $0.697 providers | $0.69 | −30% | ||
| $0.35 | $252 | $0.754 providers | $0.65 | −46% | ||
| $0.30 | $216 | $0.378 providers | $0.60 | −50% | ||
| $0.15 | $108 | $0.599 providers | $0.59 | −75% | ||
| $0.33 | $238 | $0.549 providers | $0.56 | −41% | ||
| $0.26 | $187 | $0.464 providers | $0.53 | −51% | ||
| $0.15 | $108 | $0.275 providers | $0.31 | −51% | ||
| $0.16 | $115 | $0.246 providers | $0.27 | −41% | ||
| $0.12 | $84 | $0.246 providers | $0.25 | −53% | ||
| $0.07 | $54 | $0.175 providers | $0.17 | −56% |
Cheapest = lowest in-stock on-demand listing from a provider headquartered in United States. Local median = median of those providers' own medians. Global = the index median across every provider we track.
Listings whose host reports a location inside United States — marketplace machines and hyperscaler regions. This is the section that matters for latency and data residency; the provider may be headquartered anywhere.
| GPU | Cheapest at | Where | $ / GPU / hr | Local median | Listings | Global |
|---|---|---|---|---|---|---|
| us-ashburn-1 | $8.60 | $8.60 | 1 | $8.60 | ||
| us-ashburn-1 | $15.00 | $15.00 | 1 | $7.85 | ||
| California, US | $7.38 | $8.76 | 6 | $6.59 | ||
| US | $4.61 | $7.91 | 3 | $4.45 | ||
| Washington, US | $3.61 | $3.94 | 4 | $3.72 | ||
| US | $4.18 | $11.06 | 9 | $3.50 | ||
| Virginia, US | $2.91 | $6.98 | 3 | $3.03 | ||
| eastus | $2.75 | $3.33 | 2 | $2.66 | ||
| us-ashburn-1 | $6.00 | $6.00 | 1 | $2.59 | ||
| California, US | $1.08 | $1.50 | 34 | $2.05 | ||
| us-east-1 | $1.01 | $1.83 | 8 | $1.83 | ||
| California, US | $0.43 | $1.54 | 25 | $1.79 | ||
| us-east-1 | $2.74 | $3.22 | 4 | $1.48 | ||
| us-east-1 | $1.86 | $3.50 | 9 | $1.47 | ||
| us-ashburn-1 | $2.00 | $3.23 | 4 | $1.29 | ||
| New York, US | $2.41 | $2.42 | 2 | $0.86 | ||
| Oklahoma, US | $0.69 | $0.70 | 3 | $0.82 | ||
| Utah, US | $0.32 | $1.32 | 13 | $0.80 | ||
| United States, US | $0.12 | $1.64 | 16 | $0.79 | ||
| United States, US | $0.51 | $0.61 | 18 | $0.69 | ||
| Texas, US | $0.35 | $0.74 | 19 | $0.60 | ||
| eastus | $0.53 | $0.98 | 11 | $0.59 | ||
| New Jersey, US | $0.39 | $0.67 | 3 | $0.53 | ||
| Nevada, US | $0.34 | $0.34 | 1 | $0.31 | ||
| United States, US | $0.14 | $0.16 | 11 | $0.25 | ||
| Michigan, US | $0.11 | $0.12 | 9 | $0.17 |
The whole index — 32 GPUs from every provider we track, each of which will rent to a customer in United States — with the global median and the single cheapest listing converted to USD. Tax is not included.
From a provider headquartered in United States: $1.99 per GPU-hour at io.net on 2026-09-18, about $1,433 a month at 720 hours. On a machine physically located in United States: $4.18 per GPU-hour at Vast.ai (US), 9 such listings today. Across every provider we track, the global median is $3.50 and the cheapest listing anywhere is $1.68 at Latitude.sh — any of them will sell to a customer in United States.
28 of the 48 providers we price today are headquartered in United States: RunPod Community Cloud, RunPod Secure Cloud, Vast.ai, io.net, Massed Compute, Oracle Cloud, Amazon Web Services, Modal, Akash, Beam, Cerebrium, Lambda, Microsoft Azure, CoreWeave, Lightning AI, Vultr, DigitalOcean GPU Droplets, Qubrid, Replicate, Paperspace (DigitalOcean), fal.ai, Crusoe, Baseten, SaladCloud, Thunder Compute, Together AI, GMI Cloud, TensorDock. On 31 of the 32 GPUs they price, the cheapest local listing is below the global median.
The provider's published dollar price, before tax. Providers that publish in euros are converted at the rate captured with the snapshot (2026-09-18).
For holding a model, memory per dollar: the NVIDIA RTX 3090 gives 96 GB per $1/hr. The cheapest GPU-hour of any kind is the NVIDIA RTX A4000 at $0.17. The full table above prices every GPU we track in USD; the cheapest-listing column links to who sells it.
Nearly. Most providers publish one price list regardless of where you are, so the market table on this page is the same market a buyer in any country sees — the differences are tax, currency and, for the hyperscalers, region. What does change by country is who has hardware nearby: the hosted-hardware section lists machines actually located in United States, where any exist, which is what matters for latency and data residency.