The price of intelligence,
in real time.
Flopdesk aggregates and normalizes GPU pricing and capacity across Lambda, CoreWeave, AWS, and 7+ providers — so you know exactly what compute costs, where it's available, and where the market is heading.
- 6
- providers tracked
- 8
- GPU SKUs priced
- 131
- price points / day
Last capture 9h ago
Today across the GPU market
Where compute is cheapest
On-demand $/GPU-hr by model and provider.
| GPU | AWS | Microsoft Azure | Google Cloud | Lambda | RunPod |
|---|---|---|---|---|---|
| B200 | $14.24 | · | $4.28 | $6.69 | $5.89 |
| H200 | $9.49 | $11.15 | $4.58 | $2.29 | $3.79 |
| H100 SXM | $6.88 | $12.29 | $4.20 | $3.29 | $2.89 |
| Instinct MI325X | · | · | · | · | · |
| Instinct MI300X | · | $6.55 | · | · | $1.99 |
Every major cloud, normalized
Ranked by capacity and reliability.
AWS
#1Hyperscaler with the deepest capacity and enterprise SLAs; premium pricing offset by reserved and capacity-block discounts.
Microsoft Azure
#2Hyperscaler with exclusive frontier-scale clusters and the broadest AMD Instinct availability.
Google Cloud
#3Hyperscaler pairing NVIDIA clusters with custom networking; strong for integrated data and ML pipelines.
Lambda
#4GPU cloud purpose-built for AI training and inference, popular with startups for transparent on-demand pricing.
RunPod
#5Community and secure-cloud GPU marketplace known for the lowest spot rates and per-second billing.
Vast.ai
#6Decentralized GPU marketplace aggregating spare capacity from many hosts; lowest spot rates with variable availability.
Availability is the new alpha
Track cluster availability before it moves the price.
- 1
Capacity shifts days before list prices react. We track inventory across every region so you see tightening before it hits your bill.
- 2
Availability tiers (available, limited, waitlist, sold out) are normalized across providers into one comparable signal.
- 3
Set alerts when a cluster you want frees up, or when a region you depend on starts running hot.
Built for the people moving compute
“Placeholder quote: the price tape replaced three spreadsheets and a Slack channel. We now negotiate reserved capacity with actual market context.
“Placeholder quote: availability alerts have caught two regional crunches before they hit our training runs. The capacity signal alone pays for the seat.
“Placeholder quote: we model the compute market the way we model rates. Flopdesk is the first normalized feed we have trusted.
Intelligence at every scale
From founders watching burn to funds modeling the market.
Free
Track the market.
- Live spot pricing (15-min delay)
- Market overview dashboard
- Top providers & GPU rankings
- 7-day price history
Start tracking the compute market today
Join the desks, founders, and funds using Flopdesk to price the AI economy.