Trader consensus around $5–6 per GPU-hour for H200 rentals by end-August reflects persistent demand for the 141GB HBM3e Hopper card in memory-intensive large language model inference and training, tempered by rising supply from neocloud providers and early Blackwell capacity. Recent July data show marketplace rates clustered at $3.49–$4.39 on platforms like Hyperbolic and Runpod, while hyperscaler on-demand remains higher near $5–10, creating a wide but converging band as contracts roll off and new clusters come online. Competitive dynamics hinge on whether specialist clouds can absorb spillover from constrained B200 availability without further price erosion, alongside any short-term demand spikes from AI workload scaling before September. The narrow spread between the top two buckets highlights uncertainty over how quickly additional H200 inventory will hit the spot market.