The AI Compute Radar aggregator has expanded its hourly GPU rental price tracking to include Azure and Oracle Cloud, ensuring that every listed price is tied to a specific virtual machine or bare-metal instance name.
- The update compares per-GPU-hour costs for A100, H100, H200, and B200 GPUs against Vast.ai and RunPod medians.
- Azure prices are derived from Linux pay-as-you-go VM costs divided by GPU count, while Oracle provides direct GPU-hour pricing.
- The data highlights that list prices often hide the unit of sale, such as Oracle's H100 requiring an eight-card bare-metal machine.
- Significant price gaps remain between cloud lists and marketplaces, with A100 and H100 costs being several times higher than marketplace medians.
This addition provides a more transparent view of major cloud provider pricing by explicitly linking rates to the underlying hardware configurations.