Renting Nvidia's H100 AI chip now costs about 40% more than it did six months ago, and on-demand capacity is effectively sold out across cloud providers. Buyers seeking a full cluster face waits of up to eighteen months, reversing the price declines that defined 2025.
Renting Nvidia's H100 AI chip climbed roughly 40% between October 2025 and March 2026, according to data from SemiAnalysis, with average rates rising from $1.70 to $2.35 per GPU-hour. Securing an actual cluster now means 12 to 18 months of waiting.
That marks a sharp reversal from 2025, when GPU rental prices dropped more than 60% from their 2023-2024 peaks as Neocloud specialists such as CoreWeave and Lambda expanded capacity alongside the major hyperscalers.
Demand outpaces even small deployments
Inference workloads and multi-agent AI systems have pushed compute requirements beyond what current supply can handle, and on-demand H100 capacity is effectively sold out across Neoclouds and hyperscalers alike. Even clusters as small as 8 nodes, or 64 GPUs, have become difficult to procure on short notice.
Providers pushed through 15% to 20% price increases in February 2026 alone. Rentals across the broader market now range from $2.19 to more than $4 per GPU-hour, depending on the provider and whether capacity is reserved in advance or bought on the spot.
Blackwell backlog adds pressure
Lead times for Nvidia's B200 and GB200 chips have stretched into mid-2026, with most available 2026 capacity already committed. Some operators have locked in older H100 contracts at legacy rates, and a few have extended commitments through 2028 — a four-year contract length that was nearly unheard of in cloud GPU markets 18 months ago.
By mid-2026, pricing showed some early signs of stabilizing, though levels remained well above where they stood six months earlier.
Providers gain leverage
Higher utilization and rising prices are lifting revenue per rack for GPU cloud providers. Firms that locked in long-term supply deals with Nvidia at favorable terms are now running a spread trade: cheap wholesale, expensive retail. Hyperscalers like AWS, Google Cloud, and Azure can absorb the temporary margin pressure, but smaller Neoclouds may hold more negotiating leverage than expected when every GPU is spoken for.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
