Strong AI training and inference demand continues to anchor NVIDIA B200 rental rates near the middle of observed ranges, with neocloud on-demand medians holding around $6.69 per GPU-hour and spot offerings dipping to $3.20–$4.95 while hyperscalers list up to $14–$16. Recent auction results at providers like Nebius, where customers paid 15–20% premiums for guaranteed Blackwell capacity, underscore resilient appetite that has lifted 7-day medians by roughly 6%. Supply constraints from HBM3e and advanced packaging persist even as Blackwell ramps, producing a wide spread across 18+ providers and keeping the $6.00–$7.00 band tightly contested. With resolution just weeks away, any fresh capacity announcements or demand surges from large model deployments could shift the narrow gap between the two leading buckets.