**Persistent uncertainty between expanding AI infrastructure supply and unrelenting demand for high-performance compute is keeping H100 rental prices at end-2026 tightly contested across the Polymarket ranges.** Current on-demand rates cluster between roughly $2 and $4 per hour on specialist providers, with hyperscalers still commanding premiums above $6–10, after sharp declines from 2024–2025 peaks as more capacity came online. The near-even implied probabilities reflect competing forces: continued strong enterprise and model-training demand that has recently lifted some 1-year contract rates 20–40% in early 2026, versus the accelerating availability of Blackwell B200 GPUs, which are shifting older H100 units toward mainstream workloads and could exert further downward pressure. Persistent bottlenecks in TSMC CoWoS packaging and HBM memory limit how quickly new supply can fully ease constraints, while hyperscaler buildouts and potential AMD or custom silicon alternatives add longer-term competitive dynamics. Traders appear to weigh near-term demand resilience against the historical pattern of price normalization once next-generation hardware ramps, leaving resolution dependent on 2026 capacity additions and AI workload growth.