Market Data

Per-provider list-ask time-series across hyperscalers, neoclouds, and managed-Kubernetes operators that run their own datacenters. Solid lines mark guaranteed availability; dashed lines mark interruptible. Bundled offerings are flagged and can be toggled.

Least-expensive available rates from third-party-host marketplaces. Vast.ai restricted to verified hosts; RunPod split by sub-product (Secure / Community / Spot / Cluster). MIN is the headline; Vast median sits behind it as a faded backdrop.

The quantity side of the market: a diffusion read of how many genuine-capacity venues show launchable stock — breadth (X of Y venues), two bounds (any configuration vs full 8-GPU node), charted over time so the slope reads demand tightening or easing. Measured Vast utilization (denominator-fixed rented/total) rides alongside as the depth card.

Vast is the one live, executable marketplace in the panel, so it is pulled out of the list-ask benchmark and re-surfaced here on both axes: price (how the live venue sits vs the ex-Vast Tier-3 list-ask floor) and quantity (implied utilization — machine-reconstructed share of listed capacity not currently rentable). Single-venue signal, adjacent to the methodology.

Shadeform aggregates 19 GPU clouds into a single view: where each chip is procurable, per-provider price spread, boot-time tradeoffs, config availability, long-tail catalog. Collection is paused pending the pipeline migration; the page shows the last published snapshot. Observability data, not part of the CCIR rate methodology.

The per-host order-book deep dive — distribution shape, quality bands, $/perf frontier, supply structure, geography, egress and storage. Collection is paused pending the pipeline migration; the page shows the last published snapshot. For live Vast signals see the Vast Overlay. Observability data, not part of the CCIR rate methodology.