AI compute has a price board now.
GPU Rental Prices tracks live cloud GPU rental offers across providers. The latest snapshot used here includes 305 offers from 34 providers, giving a market-style view of what different accelerators cost per GPU-hour.
The newest high-end chips sit at the top of the board. B300 rentals show a median price of $8.24 per GPU-hour, while B200 rentals show a median price of $6.59 per GPU-hour. H200, H100, MI300X, A100, L40S, and RTX-class GPUs sit lower on the curve.
The interesting part is the spread. Some chips have a tight rental range, while others show wide dispersion between the cheapest and most expensive offers. That dispersion says the GPU market is not one clean price. It is a live capacity market shaped by provider, region, availability, generation, memory, and demand.
For investors, this is useful because AI compute is becoming more observable. Instead of only watching hyperscaler capex or chipmaker revenue, public GPU rental prices offer a more direct signal for supply, scarcity, and willingness to pay.