Persistent supply constraints and resilient AI demand have anchored trader sentiment around elevated B200 rental rates, with the $7.00+ bracket holding 56.5% implied probability. Current on-demand medians near $6.69 per GPU-hour across neoclouds, alongside spot floors at $3.20–3.50 and hyperscaler premiums above $14, reflect ongoing TSMC CoWoS-L packaging limits and multi-quarter backlogs that have limited volume growth into late 2026. Recent auctions at providers like Nebius, where customers paid 15–20% premiums, underscore sustained frontier training and inference workloads outpacing supply additions. Key swing factors include further Blackwell production ramps, power availability, and any acceleration in AMD MI300X adoption before year-end resolution.