Renting an Nvidia H200 now costs roughly 160% more than it did at the start of the year. The increase covers the period from January 1, 2026 through mid-October and points to demand for AI computing power that supply still cannot match. The H200, Nvidia's Hopper-generation data center GPU, is an upgraded variant of the H100 built with faster memory. For teams that lease GPU capacity by the hour instead of buying hardware outright, those rates are the direct price of access to AI compute.

The H200 Leads a Broad Price Climb

The H200 is not the only Nvidia chip becoming more expensive to rent. It is simply rising faster than the rest of the company's lineup. Year-to-date rental increases across Nvidia's models compare as follows:

• H200: up about 160%
• B200: up about 120%
• RTX PRO 6000: up around 65%
• H100: up approximately 35%

The list spans three chip generations: the H100 and H200 belong to the Hopper line, the B200 is the lead data center chip of the newer Blackwell generation, and the RTX PRO 6000 targets professional workstations.

The most unusual data point belongs to the A100, the Ampere-generation part that predates them all. Despite being six years old, the chip now rents for more than it did in January.

The H200's outsized jump is consistent with the hardware's capabilities. The chip carries 141 GB of HBM3e memory and 4.8 TB/s of memory bandwidth, giving it 1.76 times the memory capacity of its predecessor, the H100. Those two specifications determine how large an AI model a single GPU can hold and how fast it can be fed data, which is what renters are effectively buying by the hour.

What Renters Are Actually Paying

Median on-demand rental prices for the H200 sit at around $4.27 to $4.94 per GPU-hour across tracked providers. Spot marketplaces are cheaper: H200 rates there can fall as low as about $1.47 per GPU-hour, with spot pricing ranging up to $3.88. Spot listings sell spare capacity at variable rates, and the discount is substantial — the floor of the spot market sits at roughly a third of the on-demand median.

At the other end of the market are the hyperscalers, the giant cloud platforms such as AWS and Azure. Their H200 rates average from about $7.91 to more than $10.60 per hour, meaning an identical GPU-hour can cost roughly seven times more at the top of that range than the cheapest spot rate.

Individual providers are also nudging prices upward. Nebius raised its on-demand H200 price from $4.50 to $5.40 per GPU-hour, a 20% increase that took effect on October 1, 2026. The company did not stop there. In the same adjustment, Nebius raised H100 rates by 17%. The change landed inside the window the rental data covers, so part of the year-to-date climb already reflects explicit provider rate decisions — moves that surface first on individual providers' rate cards before they reach the tracked medians.