What GPU Compute Price Data is Revealing

Artificial Intelligence ("AI") is scaling across industries at an unprecedented rate and the cost of compute has become a material line item on company balance sheets. True hedgers, such as frontier AI labs training large language models, and neoclouds powering AI-driven products, are all exposed to GPU rental price risks.

American Innovation Exchange LLC ("AI Exchange"), a subsidiary of Architect Financial Technologies Inc., in preparation to launch U.S. regulated compute derivatives contracts looked at Compute Desk's daily rental prices for chips such as Nvidia H100s, H200s, B200s, or B300s from the past couple of months. While there tends to be more volatility at a chip's launch because supply is limited to a few suppliers (you can see this in the beginning of the timeseries for B200 and B300 chips) overall the chips told a similar story. We saw chips move through a long slide down through most of 2024 and 2025, a trough late last year, and a sustained rally that's still running today.

Diving Into the Data

From Compute Desk market data, we saw every chip in the dataset bottom out within a few months of each other, ending in December 2025. Each chip bounced back but the timing is important because this happened across the board. Also, this isn't one chip getting cheaper or more expensive on its own. We saw a market-wide swing in compute pricing, hitting older and newer hardware at roughly the same time.

H100, the chip with the longest price history here, tells the fullest version of the story. Compute Desk started tracking the chip in April 2024 at a price of $4.57/hr. It slid almost continuously for a year and a half, bottoming at $1.87/hr on December 30, 2025. This is a 59% drop from its high. Since then it's clawed back to $2.57/hr, up 12% over the last 90 days, with 6 points of that gain coming in just the last month.

H200 followed the same arc but on a shorter timeline. It entered the dataset in October 2024, bottomed at $2.37/hr on September 27, 2025, and has since rallied 16% to $3.01/hr.

B200 is the most volatile chip in the set. Its price has swung by nearly $0.67/hr in both directions over its history. It peaked at $5.78/hr in June 2025, cratered to $3.76/hr by late November, and has since staged the sharpest recovery of the group. It is up 21% in 90 days, including a 9% jump in just the last 30.

B300, the newest chip, never really had a "before" period in this dataset. It entered in November 2025 already near the bottom of the broader market cycle and has climbed 25% since. However that climb has flattened out over the past month after peaking near $5.26/hr in late April.

What does this mean?

There is no surprise in pricing trends if you look at the chips side by side. The new, more capable chips are priced at a premium. However, the more interesting signal is the timing. A coordinated bottom across four different chip generations in a two-to-three month window highlights systematic challenges such as tighter supply, a demand surge, or both. The data may be correlated with the market that is growing around AI commodities and products to manage the risk.

AI Exchange is preparing to launch U.S. regulated Compute Futures (pending regulatory approval), a derivatives contract that provides price transparency and risk management to companies with exposure to GPU price volatility. Modeled on the same market infrastructure as oil and gas, this is a landmark moment for the new compute markets. This new corner of finance is quietly taking shape at the intersection of AI and commodities trading, and we are pleased to be a part of the growth.