**Persistent high demand for Blackwell B200 GPUs, driven by frontier AI training and high-throughput inference workloads, continues to support elevated rental rates into late 2026.** Median on-demand prices hover near $8 per GPU-hour across providers like CoreWeave, Lambda, and RunPod, with spot rates as low as $4–$4.50 and peaks above $9–$14 at hyperscalers. The B200’s 180 GB HBM3e memory and 8 TB/s bandwidth deliver clear advantages over H200 or H100 for memory-intensive large language models, sustaining premiums even as more capacity comes online. Recent provider price increases (e.g., Nebius in October) and major deals such as Tencent-Oracle leasing reflect tight HBM supply and strong AI-lab uptake. Expanding neocloud and marketplace availability could ease rates later in the year, but ongoing production constraints and next-gen demand keep near-term sentiment firm. Traders watch spot vs. reserved spreads and new cluster launches for resolution signals.