Compute
GPU prices, inference costs, and industry news.
Current prices
Median on-demand rate · Updated September 28, 2026
A100
$1.60
per GPU-hour
H100
$3.36
per GPU-hour
H200
$4.46
per GPU-hour
B200
$6.89
per GPU-hour
Inference calculator
Compare example GPU setups for open models and estimate rental costs at the current H100 median rate.
Example setup · one replica
1× H100
80 GB memory per GPU
~16 GB
Model weights
Fits on one H100. Smaller GPUs can be a better fit for light workloads.
BF16 · 2 bytes per parameter
H100 hourly rate
$3.36market median
$2,452.80/ month
1 GPU × $3.36 × 730 hours
Estimated GPU rental cost, including idle hours.
Sizing assumptions
These are example configurations, not benchmarked deployments. Weight memory is parameters × bytes per parameter; the runtime and conversation cache need additional memory. Multi-GPU pricing and availability depend on the provider.
Memory fit does not predict requests per second. Benchmark your context lengths and concurrency before choosing hardware.
Benchmark your model with realistic prompt lengths, output lengths, concurrency, and latency targets. Add replicas for peak demand and availability. At low or irregular volume, compare a model API with paying for always-on GPUs.
DeepSeek-V3 model card & reference deploymentGPU rental prices
Median on-demand rate · USD per GPU-hour
A100
-0.3%
$1.60 → $1.59
Start → current
H100
+16.3%
$2.89 → $3.36
Start → current
H200
+17.9%
$3.78 → $4.46
Start → current
B200
+10.4%
$6.24 → $6.89
Start → current
Updated September 28, 2026 · Data: GetDeploying · CC BY 4.0
What a compute market needs
A benchmark
A neutral reference for what a GPU hour is worth across providers and regions.
Available capacity
Clear terms for when capacity is available, for how long, and with what performance.
A way to commit
Buyers and suppliers need simple contracts for spot capacity, reservations, and longer commitments.
Market signals
October 2026
Oct 01, 2026 / Index
GPU prices are becoming market data
Silicon Data and Ornn publish reference indices for rental GPU prices, creating a shared language for compute costs.
Read announcementSep 17, 2026 / Capacity
CoreWeave reports higher contract pricing
The company says it signed new three-to-six-month capacity contracts at higher prices in Q3.
Read announcementApr 09, 2026 / Futures
Compute futures are moving toward launch
CME Group and Silicon Data announced plans for futures tied to GPU rental indices, pending regulatory review.
Read announcementBuilding a compute market
We’re developing a marketplace for GPU capacity. Talk to us about buying or supplying compute.