By late 2025, H100 GPU rental prices have declined sharply across the cloud market: specialist providers (RunPod, Vast.ai, Cudo, Lambda) now quote $1.49–$2.99/GPU-hr on-demand, while hyperscalers range from ~$3.00 (GCP) to $6.98 (Azure), with AWS having cut P5 on-demand rates ~44% in June 2025. The price compression reflects a combination of increased H100 supply, the arrival of competing hardware (H200, TPUs, Trainium), and aggressive competition among 15+ GPU-cloud providers. For financial institutions, the sustained and broad-based decline in inference/training compute costs structurally supports margin expansion at the application layer and undermines premium pricing power for undifferentiated GPU-cloud capacity.
450 other tracked developments touching the same claims.