GPU cloud pricing data: H100, H200, B200 and A100 hourly rental price dataset, refreshed daily
Daily GPU hourly rental price observations across AWS, GCP, Azure, CoreWeave, Lambda, RunPod, Vast.ai, Together, Nebius, Crusoe, Oracle and DigitalOcean, covering H100, B200…

paulh
@Jason build this dataset: hourly rental prices for H100, H200, B200, A100, L40S and MI300X GPUs from 12 clouds, read from their public pricing pages every day. On-demand, spot, commitment and capacity-block prices. Export to CSV or Google Sheets. Refreshed daily.

JasonApp
On it. Reading
amazon.com
google.com
vast.ai and 9 more.

JasonApp
6,286 prices, 22 columns in this dataset.
6,286 prices, refreshed daily
Sample · 50 rows
You’re viewing a sample of this dataset.
| Row# | Open row | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | A100 | A100 | Serverless inference | Global | Global | Serverless | Not stated | USD | Gpu hour | Not stated | Region unspecified on this page. | $2.72 | RunPod | 2026-10-30 | Published list | 2.72 | 2.72 | — | — | View row | ||
| 2 | H200 | NVIDIA H200 141GB HGX | NVIDIA H200 141GB HGX | Global | Global | On demand | No fixed term | USD | Gpu hour | Not stated | — | $4.29 | Crusoe | 2026-10-31 | Published list | 4.29 | 4.29 | — | — | View row | ||
| 3 | B200 | NVIDIA HGX B200 | NVIDIA HGX B200 | Europe | Regional | On demand | No fixed term | USD | Instance hour | Not stated | CoreWeave Europe table; instance-hour total. | $68.80 | CoreWeave | 2026-11-06 | Published list | 68.8 | 8.6 | 68.8 | 8 | View row | ||
| 4 | H200 | H200 NVL | Marketplace offer 58416011 | New Mexico, US | US | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 4.002222222222223 | Vast.ai | 2026-11-17 | Published list | 4.002 | 4.002 | 4.002 | 1 | View row | ||
| 5 | A100 | NVIDIA A100 | NVIDIA A100 | Europe | Regional | On demand | No fixed term | USD | Instance hour | Not stated | CoreWeave Europe table; instance-hour total. | $21.60 | CoreWeave | 2026-11-05 | Published list | 21.6 | 2.7 | 21.6 | 8 | View row | ||
| 6 | B200 | NVIDIA HGX B200 | GPU cluster | Global | Global | Spot | Interruptible, no fixed term | USD | Gpu hour | Not stated | Preemptible rate; reservation duration not stated. | $4.09 | Together | 2026-10-31 | Published list | 4.09 | 4.09 | — | — | View row | ||
| 7 | H100 | Nvidia H100 | a3-highgpu-8g | Provider list (region unspecified) | Global | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $52.700460712 / 1 hour | GCP | 2026-11-11 | Published list | 52.7 | 6.588 | 52.7 | 8 | View row | ||
| 8 | H100 | H100 PCIE | Marketplace offer 56885479 | South Africa, ZA | ZA | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 6.055014177777778 | Vast.ai | 2026-11-15 | Published list | 6.055 | 3.028 | 6.055 | 2 | View row | ||
| 9 | H200 | H200 | p5e.48xlarge | US West (Oregon) | US | Capacity block | Not stated | USD | Instance hour | Not stated | AWS EC2 Capacity Block rate; operating system fee may apply. | $54.924 USD | AWS | 2026-11-16 | Published list | 54.924 | 6.866 | 54.924 | 8 | View row | ||
| 10 | L40S | L40S | Marketplace offer 60498276 | Spain, ES | ES | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 1.0696296296296297 | Vast.ai | 2026-11-15 | Published list | 1.07 | 0.535 | 1.07 | 2 | View row | ||
| 11 | B200 | NVIDIA HGX B200 | 1-Click Cluster | Global | Global | Commitment | 2 weeks – 1 year; 256+ GPUs | USD | Gpu hour | Plus applicable sales tax/VAT/GST. | Cluster offer; GPU count is 256+. | $8.87 | Lambda | 2026-11-06 | Published list | 8.87 | 8.87 | — | 256 | View row | ||
| 12 | B200 | Nvidia B200 | a4-highgpu-8g | Provider list (region unspecified) | Global | Commitment | 3 years | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $56.7072 / 1 hour | GCP | 2026-11-10 | Published list | 56.707 | 7.088 | 56.707 | 8 | View row | ||
| 13 | A100 | Nvidia A100 | a2-ultragpu-4g | Provider list (region unspecified) | Global | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $12.014099178 / 1 hour | GCP | 2026-11-18 | Published list | 12.014 | 3.004 | 12.014 | 4 | View row | ||
| 14 | H100 | Nvidia H100 | a3-megagpu-8g | Provider list (region unspecified) | Global | Calendar mode | DWS Calendar Mode reservation | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $44.0000 / 1 hour | GCP | 2026-11-13 | Published list | 44 | 5.5 | 44 | 8 | View row | ||
| 15 | A100 | NVIDIA A100 80GB SXM | NVIDIA A100 80GB SXM | Global | Global | On demand | No fixed term | USD | Gpu hour | Not stated | — | $2.30 | Crusoe | 2026-11-21 | Published list | 2.3 | 2.3 | — | — | View row | ||
| 16 | H100 | Nvidia H100 | a3-highgpu-8g | Provider list (region unspecified) | Global | Commitment | 3 years | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $38.864383195 / 1 hour | GCP | 2026-11-19 | Published list | 38.864 | 4.858 | 38.864 | 8 | View row | ||
| 17 | A100 | Nvidia A100 | a2-highgpu-2g | Provider list (region unspecified) | Global | Flex start | DWS Flex-start scheduling | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $4.0000 / 1 hour | GCP | 2026-11-07 | Published list | 4 | 2 | 4 | 2 | View row | ||
| 18 | A100 | Nvidia A100 | a2-highgpu-4g | Provider list (region unspecified) | Global | On demand | No fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $14.69354 / 1 hour | GCP | 2026-11-10 | Published list | 14.694 | 3.673 | 14.694 | 4 | View row | ||
| 19 | A100 | Nvidia A100 | a2-highgpu-2g | Provider list (region unspecified) | Global | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $3.97826 / 1 hour | GCP | 2026-11-04 | Published list | 3.978 | 1.989 | 3.978 | 2 | View row | ||
| 20 | L40S | L40S | Marketplace offer 56594814 | Spain, ES | ES | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 1.0696296296296297 | Vast.ai | 2026-11-04 | Published list | 1.07 | 0.535 | 1.07 | 2 | View row | ||
| 21 | B200 | NVIDIA HGX B200 | GPU cluster | Global | Global | Commitment | 31-90 days | USD | Gpu hour | Not stated | Reserved rate. | $7.79 | Together | 2026-10-30 | Published list | 7.79 | 7.79 | — | — | View row | ||
| 22 | B200 | B200 | Marketplace offer 59157388 | , US | US | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 12.510416666666666 | Vast.ai | 2026-11-10 | Published list | 12.51 | 12.51 | 12.51 | 1 | View row | ||
| 23 | A100 | A100 PCIE | Marketplace offer 54026809 | Czechia, CZ | CZ | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 3.228148148148148 | Vast.ai | 2026-11-10 | Published list | 3.228 | 0.807 | 3.228 | 4 | View row | ||
| 24 | B200 | NVIDIA HGX B200 | GPU cluster | Global | Global | Commitment | 3-30 days | USD | Gpu hour | Not stated | Reserved rate. | $7.99 | Together | 2026-11-11 | Published list | 7.99 | 7.99 | — | — | View row | ||
| 25 | H100 | H100 | p5.4xlarge | US East (Ohio) | US | Capacity block | Not stated | USD | Instance hour | Not stated | Capacity Block effective hourly rate per instance. | $5.970 | AWS | 2026-11-13 | Published list | 5.97 | 5.97 | 5.97 | 1 | View row | ||
| 26 | A100 | NVIDIA A100 | NVIDIA A100 | North America | Regional | On demand | No fixed term | USD | Instance hour | Not stated | North America regional list. | $21.60 | CoreWeave | 2026-11-12 | Published list | 21.6 | 2.7 | 21.6 | 8 | View row | ||
| 27 | B200 | NVIDIA HGX B200 | 1-Click Cluster | Global | Global | Commitment | 2 weeks – 1 year; minimum 16 GPUs | USD | Gpu hour | Plus applicable sales tax/VAT/GST | Cluster offer; price per GPU per hour. | $9.86 | Lambda | 2026-10-31 | Published list | 9.86 | 9.86 | — | 16 | View row | ||
| 28 | H200 | NVIDIA HGX H200 | GPU cluster | Global | Global | Commitment | 3-30 days | USD | Gpu hour | Not stated | Reserved rate. | $4.99 | Together | 2026-11-17 | Published list | 4.99 | 4.99 | — | — | View row | ||
| 29 | H100 | H100 SXM | Standard_ND96isr_H100_v5 | Eastus | US | On demand | No fixed term | USD | Instance hour | Not stated | Linux retail list; disks/network/tax may be extra; not live capacity. Effective 2023-12-01T00:00:00Z. H100 counts from Microsoft Learn NCads H100 v5 and ND H100 v5 specifications. | 98.32 | Azure | 2026-11-12 | Published list | 98.32 | 12.29 | 98.32 | 8 | View row | ||
| 30 | H100 | NVIDIA H100 80GB HGX | NVIDIA H100 80GB HGX | Global | Global | On demand | No fixed term | USD | Gpu hour | Not stated | — | $3.90 | Crusoe | 2026-11-01 | Published list | 3.9 | 3.9 | — | — | View row | ||
| 31 | B200 | B200 | Serverless | Global | Global | Serverless | Not stated | USD | Gpu hour | Not stated | Serverless inference worker pricing. | $10.79 | RunPod | 2026-11-18 | Published list | 10.79 | 10.79 | — | — | View row | ||
| 32 | A100 | Nvidia A100 | a2-ultragpu-4g | Provider list (region unspecified) | Global | On demand | No fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $20.275191562 / 1 hour | GCP | 2026-11-04 | Published list | 20.275 | 5.069 | 20.275 | 4 | View row | ||
| 33 | H100 | H100 SXM | Marketplace offer 58248855 | Germany, DE | DE | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 2.202222222222222 | Vast.ai | 2026-11-10 | Published list | 2.202 | 2.202 | 2.202 | 1 | View row | ||
| 34 | H200 | NVIDIA HGX H200 | NVIDIA HGX H200 | North America | Regional | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | CoreWeave North America regional list. | $20.93 | CoreWeave | 2026-10-31 | Published list | 20.93 | 2.616 | 20.93 | 8 | View row | ||
| 35 | H100 | H100 NVL | Standard_NC40ads_H100_v5 | Uksouth | UK | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Linux retail list; disks/network/tax may be extra; not live capacity. Effective 2026-09-01T00:00:00Z. H100 counts from Microsoft Learn NCads H100 v5 and ND H100 v5 specifications. | 1.61238 | Azure | 2026-11-01 | Published list | 1.612 | 1.612 | 1.612 | 1 | View row | ||
| 36 | A100 | A100 SXM | Clusters | Global | Global | On demand | No commitments | USD | Gpu hour | Not stated | Multi-GPU cluster offer; price explicitly per GPU-hour. | $2.15 | RunPod | 2026-11-15 | Published list | 2.15 | 2.15 | — | — | View row | ||
| 37 | A100 | NVIDIA HGX A100 40 GB | 8x NVIDIA HGX A100 | Global | Global | On demand | No fixed term | USD | Gpu hour | Plus applicable sales tax/VAT/GST. | Instance prices take effect November 16, 2026, at 00:00 UTC; until then, log in to Lambda Cloud to see current pricing. | $1.99 | Lambda | 2026-11-09 | Published list | 1.99 | 1.99 | — | 8 | View row | ||
| 38 | H200 | NVIDIA H200 141GB HGX | NVIDIA H200 141GB HGX | Global | Global | On demand | No fixed term | USD | Gpu hour | Not stated | — | $4.29 | Crusoe | 2026-11-08 | Published list | 4.29 | 4.29 | — | — | View row | ||
| 39 | H100 | NVIDIA HGX H100 | NVIDIA HGX H100 | North America | Regional | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Spot GPU instance; North America regional list. | $19.71 | CoreWeave | 2026-11-17 | Published list | 19.71 | 2.464 | 19.71 | 8 | View row | ||
| 40 | A100 | Nvidia A100 | a2-highgpu-1g | Provider list (region unspecified) | Global | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $2.188015 / 1 hour | GCP | 2026-11-12 | Published list | 2.188 | 2.188 | 2.188 | 1 | View row | ||
| 41 | H200 | Nvidia H200 | a3-ultragpu-8g | Provider list (region unspecified) | Global | On demand | No fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $84.806908493 / 1 hour | GCP | 2026-11-06 | Published list | 84.807 | 10.601 | 84.807 | 8 | View row | ||
| 42 | H100 | Nvidia H100 | a3-highgpu-8g | Provider list (region unspecified) | Global | Calendar mode | DWS Calendar Mode reservation | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $41.6000 / 1 hour | GCP | 2026-11-07 | Published list | 41.6 | 5.2 | 41.6 | 8 | View row | ||
| 43 | H200 | H200 NVL | Marketplace offer 59966646 | Louisiana, US | US | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 41.611111111111114 | Vast.ai | 2026-11-15 | Published list | 41.611 | 5.201 | 41.611 | 8 | View row | ||
| 44 | H200 | H200 NVL | Marketplace offer 58559583 | Colorado, US | US | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 4.122222222222222 | Vast.ai | 2026-11-17 | Published list | 4.122 | 4.122 | 4.122 | 1 | View row | ||
| 45 | H100 | NVIDIA H100 80GB HGX | NVIDIA H100 80GB HGX | Global | Global | Spot | Interruptible, no fixed term | USD | Gpu hour | Not stated | — | $4.50 | Crusoe | 2026-11-14 | Published list | 4.5 | 4.5 | — | — | View row | ||
| 46 | H200 | NVIDIA HGX H200 | HGX H200 | Europe | Regional | On demand | No fixed term | USD | Instance hour | Not stated | On-demand GPU instance; Europe regional pricing. | $50.44 | CoreWeave | 2026-11-15 | Published list | 50.44 | 6.305 | 50.44 | 8 | View row | ||
| 47 | A100 | A100 PCIE | Marketplace offer 60753759 | Japan, JP | JP | On demand | No fixed term | USD | Instance hour | Not stated | Verified rentable marketplace reference panel, up to 100 offers, not exhaustive inventory. Total hourly quote; storage/network charges and live availability can vary. | 1.5488888888888888 | Vast.ai | 2026-11-17 | Published list | 1.549 | 0.387 | 1.549 | 4 | View row | ||
| 48 | H200 | NVIDIA H200 141GB HGX | NVIDIA H200 141GB HGX | Global | Global | On demand | No fixed term | USD | Gpu hour | Not stated | — | $4.29 | Crusoe | 2026-11-09 | Published list | 4.29 | 4.29 | — | — | View row | ||
| 49 | A100 | Nvidia A100 | a2-highgpu-4g | Provider list (region unspecified) | Global | Spot | Interruptible, no fixed term | USD | Instance hour | Not stated | Official A2/A3/A4 machine table has no region selector. Nearby region controls belong to a separate G4/G2 table; this is not a verified London or Iowa quote. VM total includes predefined CPU, memory and bundled local SSD; list rate is not live capacity. | $8.37526 / 1 hour | GCP | 2026-11-01 | Published list | 8.375 | 2.094 | 8.375 | 4 | View row | ||
| 50 | A100 | NVIDIA HGX A100, 80 GB | NVIDIA HGX A100 | Global | Global | On demand | No fixed term | USD | Gpu hour | Plus applicable sales tax/VAT/GST. | Self-serve, first-come access. Price takes effect November 16, 2026, at 00:00 UTC; until then, log in to see current pricing. | $3.19 | Lambda | 2026-11-07 | Published list | 3.19 | 3.19 | — | — | View row |
About this dataset
One row per listed GPU price per day, from the public pricing pages of 12 GPU clouds: AWS, Azure, Google Cloud, Oracle, CoreWeave, Lambda, Together, Crusoe, Nebius, DigitalOcean, RunPod and Vast.ai. The 21 Nov 2026 capture has 294 prices; the history goes back to 30 Oct 2026. Each row has GPU model and variant, instance, region, purchase model (on-demand, spot, commitment, capacity block), commitment, price per GPU-hour and per instance-hour, GPU count, offer notes and the exact text the price was read from. On 21 Nov the on-demand list price of an H100 ran from $3.90 per GPU-hour on Crusoe to $11.06 on Google Cloud.
What you can use it for
ML teams choosing where to rent GPUs; finance and procurement teams budgeting compute; analysts and investors tracking GPU rental prices over time; GPU marketplaces benchmarking their own rates. Is there an API for GPU cloud prices? Some clouds have pricing APIs, most GPU clouds only publish a pricing page, and none of them keep a history. This dataset reads all 12 daily and keeps every day. How do I export GPU rental prices to CSV? Sign up free to export to CSV, Excel or JSON, or send the data to Google Sheets. GPU price tracker vs dataset: this is the tracker's data itself, one table across clouds, refreshed daily and repaired when a pricing page changes.
Coverage
12 providers, 6 GPU models (H100, H200, B200, A100, L40S, MI300X), daily since 30 Oct 2026. Prices in USD as published.
Delivery options
How Jsonify collected this data
Nobody wrote a scraper for this. Someone described the data they wanted, in a sentence, and Jason, Jsonify's AI data engineer, took it from there.
The illustration shows this pipeline’s saved steps and published data.
This pipeline refreshes daily.

paulh
@Jason I need this data: Hourly rental prices for H100, H200, B200, A100, L40S and MI300X GPUs from 12 clouds, read from their public pricing pages every day. On-demand, spot, commitment and capacity-block prices. Export to CSV or Google Sheets.
Start with amazon.com, google.com and vast.ai, then find more.

JasonApp
Pipeline steps:
- Providers
- Collect
- Validate
- Scheduled daily
Done. 6,286 prices published as v9:
A100 · A1002.72 USD
H200 · NVIDIA H200 141GB HGX4.29 USD
What Jason built
The pipeline itself, and the record of every version, repair and quality check since.
Data pipeline
- 01 · sources
Sources
amazon.com, azure.com, coreweave.com · +9 more
- 02 · input
History
Saved dated GPU observations, reconciled with fresh offers…
- 03
Providers
Select the six requested GPU providers and exact H100/B200…
- 04 · fan out
Collect
Collect official GPU list rates, preserving machine totals…
- 05
Validate
Check provider coverage and retain one dated observation per…
- 06 · output
Rentals
Daily official GPU rental prices and scoped non-listings.
6,286 prices · v9
History
InputSaved dated GPU observations, reconciled with fresh offers without duplicate daily prices.
Providers
StepSelect the six requested GPU providers and exact H100/B200 model scope.
Per input record: Collect
Fan outCollect official GPU list rates, preserving machine totals, per-GPU units, regions and purchase commitments.
Validate
StepCheck provider coverage and retain one dated observation per exact GPU offer.
Rentals
OutputDaily official GPU rental prices and scoped non-listings.
Sources
SourcesThe sites this pipeline reads on every run.
Column explanations
observation_id identifies a price observation. gpu_model and variant describe the GPU; instance identifies the machine. region and country give location details. purchase_model and commitment describe the purchase terms. currency and price_basis describe the quoted price. price, per_gpu_hour and instance_hour are numeric price fields; gpu_count is the number of GPUs when available. tax_note and offer_note record pricing qualifications. evidence_quote contains source text, and provider and source_url identify the listed provider and source. observed_at and observed_date record observation time and date. price_status describes the status of the price.
Version history
- v9
Fixed Google Cloud GPU count reading when pricing tables contain line breaks. Price coverage and saved history are preserved.
- v8
Standardized equivalent no-commitment labels in new and historical GPU observations, retaining source billing wording.
- v7
Oracle GPU prices now come directly from its official price catalog matched to its hardware table, avoiding blank page prices and including all seven relevant offers.
- v6
DigitalOcean prices now match exact GPU names directly from each price card, preventing confusion with other accelerator models.
- v5
Corrected marketplace country labels and removed unsupported Google Cloud regional assignments in recurring collection and saved-history reconciliation.
- v4
Moved daily duplicate cleanup into every refresh, preserving saved history and normalizing equivalent numbers and commitment labels.
- v3
Corrected Google Cloud billing-column interpretation so unavailable on-demand prices cannot inherit a Flex-start rate.
- v2
Adds six GPU providers and A100, H200, L40S and MI300X coverage. Completes Azure per-GPU calculations and standardizes rental commitments while preserving daily price history.
- v1
Added daily H100/B200 rental data across six providers, including explicit US and UK cloud prices.
Repairs
- repaired
- repaired
- No change needed
- No change needed
- repaired
- No change needed
- No change needed
- repaired
- No change needed
- No change needed
Quality checks
- v9Issues found
Full-table checks covered 4,256 observations. Four RunPod rows violate the accepted commitment-normalization rule. Sampled live prices otherwise broadly agree; one Vast.ai quote has changed since collection.
- v8Inconclusive
Full-table structural checks passed, but Crusoe spot classifications conflict with the live source's future on-demand pricing. Collection-time evidence is insufficient to distinguish extraction error from source drift.
- v7Issues found
Full-table checks passed for structure, arithmetic, deduplication and history retention. Source verification found misclassified Crusoe prices, lost Lambda instance sizes and missing DigitalOcean billing qualifications.
- v6Issues found
Full-table checks passed basic integrity and normalization, but Lambda loses instance-size distinctions and historical DigitalOcean records omit required billing minimums. Crusoe term labeling and RunPod coverage also warrant review.
- v5Inconclusive
Structural and normalization checks passed across 297 rows. Live checks supported sampled prices across all 12 providers, but apparent offer-coverage gaps remain unresolved.
- v4Issues found
Found incorrect country values in 27 marketplace rows. Structural and normalization checks passed; GCP regional grounding remains unresolved.
- v3Issues found
Confirmed five duplicate daily GCP offers, violating R4. Provider/model coverage and price-normalization checks passed; live verification was partial.
- v2Issues found
Confirmed duplicate daily offers, a misclassified GCP billing tier, and missing Azure GPU normalization. All 365 rows received structural checks and key-field inspection; external verification was sampled.
- v1Inconclusive
All 107 rows passed structural and normalization checks. Live checks supported many prices across all six providers, but AWS coverage and GCP London attribution remain unresolved.
Limitations
These are published list prices, not negotiated or reserved-contract rates. Purchase terms differ between clouds (per-second billing, minimum GPU counts, cluster-only pricing), so compare purchase_model and offer_note before treating two prices as equal. Vast.ai is a marketplace: its rows are the lowest host offers seen that day. Some rows have no per-GPU-hour figure when the page gives only a starting price for a product tier. Region is recorded as the page states it; several clouds publish one global price.
Ready to start?
Frequently asked questions
Is this sample data free to use?
How do I credit Jsonify?
Can I buy the full dataset?
How is that different from collecting my own?
What do Download and Customize do?
Does collecting my own cost anything?
What does a live feed for my team include?
I run a site this data was read from. What can I do?
Need something different?
Describe it. Get verified rows, kept current.