Price discovery for AI compute, from silicon to tokens.
Compare every way to buy GPU compute — buy the card, rent it by the hour, run it serverless, or pay per token — across providers, countries and currencies.
- GPU models
- 6
- Providers
- 142
- Live offers
- 23
- Rent from
- $0.22/hr
Ways to buy
Four ways to buy GPU compute
The same compute, sold at four levels of abstraction. Pick how much you want to manage — the more you hand off, the more you pay for the convenience. Renting is live today; the rest are coming.
Buy GPUs
$/GPU
Retail & wholesale purchase prices
Best for
Steady, 24/7 workloads where you'll amortize the hardware.
Rent GPUs
$/GPU-hr
Hourly cloud GPU rentals
Best for
Training runs and full control, without the capital outlay.
Serverless GPUs
$/GPU-hr active
Pay-per-second, scale to zero
Best for
Bursty or spiky inference that should scale to zero when idle.
Tokens
$/1M tokens
Per-token model API pricing
Best for
Calling a model without running any infrastructure yourself.
Live rentals
Prices, sliced every way that matters
Cheapest overall, biggest movers, cheapest per GPU and provider type, best value per GB. Every card is a live top-5 — click through for the full list.
Coverage
142 providers, one comparison
Hyperscalers, dedicated GPU clouds and peer-to-peer marketplaces — all normalized to the same units so the prices are actually comparable.
Run a workload
Skip the tables — start from what you want to run
Describe a workload and get the cheapest way to run it, ranked across the whole stack. Renting is live today; the other options join as data lands.
e.g. serve Llama-3-70B · 20M tokens/day
- TokensBest$2,140/mo
- Serverless$2,880/mo
- Rent 1×H100$3,300/mo
illustrative — live figures come from your inputs