Renting Nvidia's B200 GPU has become markedly more expensive over the past quarter. Rental rates for the Blackwell-series chip climbed 74% over the past three months, according to a GPU rental marketplace, as demand for artificial intelligence compute continues to run ahead of supply across multiple pricing trackers.

Independent research points in the same direction. Research firm SemiAnalysis recorded a 79% increase in B200 on-demand rental rates, bringing the price to $8.01 per GPU-hour by the end of September 2026.

The Ornn Compute Price Index tells a similar story over a slightly later window. Between late July and late October 2026, the index showed the B200 rental price rising from $6.58 to $9.76 per GPU-hour, a 48% increase.

A fragmented market with a wide price spread

Live listings from early November 2026 underscore just how fragmented the market remains. Across roughly 10 to 20 cloud service providers, on-demand B200 rates ranged from $1.89 to $16.11 per GPU-hour.

The top of that range belongs to the biggest names. Major providers such as Google Cloud and AWS listed rates exceeding $14 to $16 per GPU-hour, while smaller players priced far lower.

Most activity sits in the middle of the market. Typical on-demand medians landed between $6 and $8.50 per GPU-hour.

Spot pricing offers a discount for buyers willing to accept some uncertainty. Where available, spot rates often fell in the $3 to $5 range. Spot capacity is cheaper because it can be reclaimed or interrupted, making it suitable for workloads that can tolerate a pause.

Why the B200 is in such demand

The B200 is part of Nvidia's Blackwell series, built for large-scale AI training and inference. The chip carries 180 GB of HBM3e memory, the high-bandwidth memory that allows large models to move data quickly enough to stay useful.

Demand for Blackwell-series chips has consistently outpaced supply. The broader industry shift from traditional GPUs toward advanced architectures such as Blackwell has also created friction across cloud computing: providers need time to acquire, install, and bring new hardware online, and demand has repeatedly moved faster than that buildout.

No quick relief on the horizon

The supply picture does not appear likely to improve soon. Nvidia's own forecasting points to limited availability stretching through 2028.

Component costs add another layer of pressure. Rising prices for parts, including memory, are expected to feed into the cost of AI-capable infrastructure as the industry prepares for demand to continue into 2027.

What it means for AI builders and their backers

The most direct impact falls on companies that rent rather than own their compute. Startups and mid-sized AI firms often rely on cloud GPUs because buying hardware outright requires large upfront capital. For them, a jump in hourly rates flows straight into operating costs.

The wide price spread across providers also creates a strategic decision. Companies willing to work with smaller providers, or to design workloads around interruptible spot capacity, may capture meaningful savings compared with paying premium rates at the largest clouds.

Volatility is the other factor to watch. With supply constraints projected to persist and different trackers already showing gains ranging from 48% to 79% depending on the window, rental prices may continue to swing. The key signals ahead are the next readings from SemiAnalysis, the Ornn Compute Price Index, and live provider listings. If medians keep drifting above the current $6 to $8.50 band, pressure on AI budgets will intensify.

Disclosure: The original article was edited by Diego Almada Lopez.