**Current B200 rental rates cluster between roughly $3.50 and $7 per GPU-hour on most neocloud and marketplace providers, with a median on-demand price near $6–6.80 and hyperscaler quotes running higher.** Trader consensus on the Polymarket, which favors the $6.50+ bracket at 38.5% followed by the $6.00–6.50 range at 21%, reflects ongoing supply constraints from TSMC CoWoS-L packaging capacity and HBM3e memory allocation that continue to limit rapid volume growth even as more Blackwell clusters come online through late 2026. Spot and interruptible rates have traded as low as $3.20–3.95 in recent weeks on platforms such as Vultr and Prime Intellect, yet sustained demand from frontier model training and inference workloads has kept committed and on-demand pricing firmer, with many verified offers landing in the $5.50–8 range. The modest probability spread across the $5–6.50 bands highlights uncertainty around October delivery timelines versus any near-term acceleration in output or new large-scale AI commitments that could ease tightness. Key upcoming catalysts include additional neocloud capacity additions and any shifts in hyperscaler allocation behavior before month-end.