The cost of renting NVIDIA’s flagship AI chip is coming back down to earth. H100 GPU rental prices have declined after spiking earlier this month, according to data from Ornn AI Inc., which tracks the market through its Ornn Compute Price Index (OCPI).

The numbers behind the volatility

One-year contract rates for H100 rentals surged roughly 40% from $1.70 per hour in October 2025 to $2.35 per hour by March 2026. The spot and on-demand market has been even wilder. As of mid-May 2026, on-demand H100 rental pricing ranged between $1.38 per hour and $11.01 per hour. That spread reflects variation across providers, contract terms, and geographic regions.

Social media commentary on May 27 pointed to Ornn AI’s data as a contributing factor in the market’s recognition that the spike was fading.

GPU compute gets the Wall Street treatment

Ornn AI’s OCPI has been available on the Bloomberg Terminal since April 2, 2026. Then on May 19, the Intercontinental Exchange (ICE), one of the world’s largest operators of financial exchanges, announced plans to launch GPU compute futures contracts tied to the OCPI. These contracts would cover the H100 among other models.

Futures contracts allow companies to lock in future prices, hedging against volatility. They also let speculators make bets on where compute costs are headed.

What this means for investors

The near-term takeaway is straightforward: H100 rental costs are normalizing after the early May spike. The 40% run-up from October to March was already straining balance sheets.

The wide spread between on-demand pricing ($1.38 to $11.01 per hour) highlights that this market is still fragmented across different providers, geographies, and terms. The OCPI normalizes across these variables. When ICE’s futures contracts go live, hedge funds, commodity traders, and institutional investors will have a direct mechanism to express views on AI demand without buying equity in any single company.