Availability & Utilization
Two lenses on GPU supply. Availability: across a panel of neoclouds and GPU marketplaces we probe, how many show launchable stock, so the slope reads demand tightening or easing. Utilization: of the capacity the one marketplace we can measure lists, how much is rented, tracked over time. Counts publish as ranges.
Availability over time
Rolling 7-day count-weighted availability: the mean share of capacity-probes finding stock across genuine-capacity venues. Dashed verticals flag a change in the venue panel. Cluster (8×) vs Fractional.
How this is measured
Each source's window rate is its share of capacity-probes that found stock, count-weighted within source. Two layers, same panel. The tiles report breadth: of the venues that disclosed a signal, how many had a window rate of at least 10% of cycles (the floor keeps a venue with one available probe all week from reading as available until history rolls out of the window). The chart plots the count-weighted mean venue rate, which moves as availability changes rather than in window-edge steps. Both shown as Fractional (any shape) and Cluster (full 8-GPU node). Each rolling point uses the venues present in that window, and dashed verticals flag where the panel set changed, so added venues don't read as a demand move. Catalog-eligibility and probabilistic-score sources are excluded from the headline; resellers that duplicate a directly probed operator are deduped. Venue counts and rates publish as ranges. Method updated Jul 30, 2026 (floor + mean-rate chart). Utilization is a separate measured metric, never blended into availability.
Measured utilization one marketplace source
One GPU marketplace discloses rented against total listed. Rolling 7-day utilization, denominator fixed to each window's median listed capacity.