What an hour of compute costs today.
Every model, from B200 to RTX 4090, priced by the open market right now. The number that decides what AI costs to build, updated every ten minutes.
Every model, from B200 to RTX 4090, priced by the open market right now. The number that decides what AI costs to build, updated every ten minutes.
median $9.38 · 5 offers
median $4.61 · 9 offers
median $2.93 · 9 offers
median $0.587 · 64 offers
| GPU | Memory | Cheapest / hr | Median / hr | Median, relative | Offers | Cheapest in |
|---|---|---|---|---|---|---|
B200 Datacenter | 192 GB | $7.62 | $9.38 | 5 | Virginia, US | |
H200 Datacenter | 141 GB | $2.63 | $4.61 | 9 | Georgia, US | |
H100 SXM Datacenter | 80 GB | $1.47 | $2.93 | 9 | France, FR | |
H100 PCIE Datacenter | 80 GB | $1.87 | $2.30 | 11 | France, FR | |
A100 SXM4 Datacenter | 40–80 GB | $0.431 | $0.669 | 18 | California, US | |
L40S Datacenter | 48 GB | $0.467 | $0.801 | 18 | Japan, JP | |
RTX PRO 6000 WS Workstation | 96 GB | $1.12 | $1.41 | 37 | Colorado, US | |
RTX 5090 Consumer | 32 GB | $0.402 | $0.575 | 64 | Beijing, CN | |
RTX 4090 Consumer | 24 GB | $0.329 | $0.487 | 64 | United States, US |
B200
Datacenter · 192 GB
H200
Datacenter · 141 GB
H100 SXM
Datacenter · 80 GB
H100 PCIE
Datacenter · 80 GB
Read 1s ago from Vast.ai's public marketplace: rentable on-demand offers only, multi-GPU machines divided down to one GPU. A snapshot of one open market, not a quote. COREVANE does not rent or resell capacity.
A100 SXM4
Datacenter · 40–80 GB
L40S
Datacenter · 48 GB
RTX PRO 6000 WS
Workstation · 96 GB
RTX 5090
Consumer · 32 GB
RTX 4090
Consumer · 24 GB