**Supply ramp and intensifying competition among cloud providers are the main forces behind the dispersed market-implied odds for B200 rental rates at year-end.** Recent data show on-demand prices softening from March–May peaks near $6.11/hr to averages of roughly $4.70–$6.00/hr by mid-2026, with spot quotes as low as $2.12–$3.50 across neoclouds such as Spheron, RunPod, and Lambda. Improved TSMC 4NP yields, easing HBM3e memory supply, and broader B200 inventory at secondary providers have increased available capacity, while the upcoming Blackwell Ultra B300 is expected to shift some workloads and further compress B200 pricing toward the $2.50–$3.00 range by Q4. Persistent frontier-model training, inference demand, and hyperscaler premiums continue to support higher quotes, creating the balanced probability distribution across $3.50–$7.00+ buckets. Key swing factors include the pace of additional neocloud deployments and any late-year demand surge from agentic AI workloads.