← GPU cloud (on-demand)
CoreWeave
CoreWeave H200
H200 nodes with InfiniBand. $50.44/hr per 8-GPU HGX node = $6.31 per GPU (North America on-demand).
USD per GPU-hour
$6.31
Verified Oct 4, 2026
#17 cheapest of 20 in GPU cloud (on-demand)
Official pricing ↗Capabilities, explained
What this API can do, and what each capability means for your project.
GPU
H200
The GPU model. Newer generations are faster per dollar for large models.
VRAM
141 GB
Memory per GPU. A 70B-parameter model needs ~140 GB in 16-bit, or ~40 GB quantised.
Interconnect
infiniband
How fast GPUs talk to each other; matters for multi-GPU training.
Regions
us, eu
Where the hardware is located (latency, data residency).
Per-second billing
No
You pay only for the seconds used rather than full hours.
Price history
Only one price recorded so far. The chart appears once the price changes.