Nvidia H100 GPU Rental Costs Jump Nearly 40% Over Six Months: SemiAnalysis

Rental contract pricing for Nvidia's H100 GPUs has climbed nearly 40% over a six-month period, reaching $2.35 per hour per GPU in March from a low of $1.70 in October 2025, according to a recent analysis.

A newly released price index for one-year H100 rental contracts, compiled from monthly surveys of more than 100 market participants, highlights this significant price increase.

On-demand GPU rental capacity is reportedly sold out across all GPU types. The analysis indicates that customers who have secured on-demand instances are holding onto them tightly, unwilling to release capacity back into the market even as prices rise.

Evidence of intense demand includes customers competing to pay high rates for spot instances, some major providers ceasing sales of single nodes, and H100 contracts being renewed at original rates from years ago, some extending through 2028. Sourcing even small clusters of H100 or H200 GPUs has become difficult, with many providers reporting they are completely sold out and have no Hopper-series GPUs becoming available.

The report also notes that availability for Nvidia's newer Blackwell GPUs is extremely constrained. Lead times for new Blackwell deployments are now stretching into mid-2026, driven by strong demand for open-weight AI models and a significant surge in inference workloads.

Prior to late 2025, the industry widely expected rental prices for Hopper chips like the H100 to fall as Blackwell deployments increased, due to the latter's superior compute efficiency. Contrary to these expectations, demand for H100 chips remained strong and even intensified towards the end of 2025.

A major driver of demand in early 2026 originated from native media generation applications. Platforms for generating and refining images and video at scale are creating massive increases in computational demand. Usage of other advanced AI models has also seen substantial growth.

Despite clear signals of tightening supply and rising prices—conditions that typically benefit cloud providers through improved margins—public market sentiment toward several major providers has turned negative, with their share prices declining.

The analysis suggests the market remains focused on a longer-term narrative of eventual oversupply and commoditization, which has overshadowed the current reality of sustained scarcity and pricing power for providers.

Looking ahead, the report identifies three key factors that will influence whether GPU rental prices remain high: the pace at which new Blackwell clusters come online and their impact on the compute shortage; the potential worsening of the semiconductor supply shortage; and the continued growth in recurring revenue and token consumption from AI labs.

For now, the clearest beneficiaries are providers with shorter-duration contracts that can reprice quickly, large existing installations of H100 GPUs, and near-term plans to add new capacity. These providers are positioned to benefit from immediate margin expansion, while those securing next-generation capacity under long-term contracts are also expected to gain.