**Strong AI inference and small-scale training demand continues to compete with gamers for limited RTX 5090 supply, sustaining elevated rental rates on platforms like Vast.ai, RunPod, and Salad.** The Blackwell flagship’s 32 GB GDDR7 memory and high tensor performance make it an attractive, lower-cost substitute for datacenter accelerators, while GDDR7 allocation constraints and Nvidia’s prioritization of professional SKUs have kept retail and secondary-market availability tight since mid-2026. Recent reports confirm ongoing bulk purchases of consumer cards for AI servers, with next-gen RTX 60-series hardware now delayed until 2028. As of early October, on-demand rates cluster around $0.35–$1.08 per GPU-hour (medians near $0.60–$0.75), producing the closely matched market probabilities. Short-term catalysts include potential holiday restocks, spot-capacity fluctuations, and any shifts in AI workload intensity that could push rates toward the $0.55–$0.70 bands or higher.