GridStackHub's pricing database is the foundation of all our comparison tools, calculators, and forecasts. This page explains exactly how we collect, validate, and publish GPU pricing data so you can evaluate its reliability for your use case.
Data Collection
We collect GPU pricing from three primary sources:
- Provider public pricing pages: The primary source for all providers. We scrape and parse pricing from each provider's official pricing page on a daily schedule. Source URLs are stored with each data point for verification.
- Provider pricing APIs: Where providers offer machine-readable pricing APIs (AWS Price List API, GCP Cloud Billing, etc.), we query these directly. API responses are considered authoritative for those providers.
- GPU marketplaces: Spot and interruptible pricing on Vast.ai, RunPod, Salad, and similar platforms is collected via their APIs in real time. Marketplace prices fluctuate with supply and demand.
Price Normalization
Direct provider price comparison is harder than it looks. Providers package GPU compute differently:
- AWS sells H100 instances with 8 GPUs per instance — we divide by 8 to get per-GPU rate
- CoreWeave sells single-GPU instances — used directly
- Some providers bundle storage or networking in compute prices — we document this in instance notes
- Reserved pricing is expressed as effective hourly rate (total commitment / hours committed)
All prices in our database are normalized to: USD per GPU per hour, excluding storage and egress.
Data Freshness
| Data Type | Update Frequency | Lag Tolerance |
|---|---|---|
| On-demand pricing | Daily | < 48 hours |
| Reserved pricing | Daily | < 7 days |
| Spot / marketplace pricing | Real-time (API) | < 1 hour |
| Availability (GPU in stock) | Daily | < 48 hours |
| Historical snapshots | Daily archive | — |
Accuracy Standards
Our accuracy commitment:
- Prices are verified against provider source pages before publishing
- Price discrepancies > 5% between our data and provider pages trigger an immediate re-scrape
- Each pricing record includes a
last_verifiedtimestamp - Prices older than 7 days are flagged as "may be stale" in our UI
- We do not estimate or interpolate prices — if we can't verify a price, we remove the record
Provider Coverage
We currently track 32+ GPU cloud providers including:
- Hyperscalers: AWS, Google Cloud Platform, Microsoft Azure, Oracle Cloud Infrastructure
- GPU-specialized: CoreWeave, Lambda Labs, RunPod, Vast.ai, Crusoe, Together AI, Fluidstack, Voltage Park
- Mid-tier: Paperspace (by DigitalOcean), Coreweave, Tensordock, DataCrunch, Oblivus
- Regional: OVHcloud, Hetzner, Scaleway, Exoscale (European providers)
- New entrants: Nebius (Yandex Cloud), Salad, JarvisLabs, Prime Intellect
View the full live database: GPU Pricing Research Database →
Frequently Asked Questions
See the live pricing database
All data, all providers, all GPU models. Filter and compare in seconds.
Open Research Database →