Punters on Polymarket are wagering on what it will cost to rent Nvidia's H100 chip by the end of July, and the market says more about the AI economy than almost any equity index.
Traders currently assign a 43% probability that the benchmark rental rate lands between $2.30 and $2.60 per hour, with a 26% chance of $2.60 to $2.90.
Only 2% expect prices below $2.00, the level that would signal genuine oversupply.
The contract resolves against the Ornn H100 Index, a benchmark tracking hourly rental rates across cloud providers.
That such an index exists at all may be the real story: compute is becoming a tradeable commodity, like oil or wheat.
The Ornn index has been available on the Bloomberg Terminal since April, and Intercontinental Exchange, one of the world's largest exchange operators, has announced plans to launch GPU futures contracts tied to a compute price benchmark.
That would give hedge funds and commodity traders a direct way to bet on AI demand without buying shares in Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) or Microsoft Corp (NASDAQ:MSFT).
The price itself is a live macro signal. One-year H100 rental contracts surged roughly 40% between October 2025 and March 2026, from $1.70 to $2.35 per hour, as an unexpected compute crunch took hold.
Most analysts had assumed the opposite: that older Hopper chips would tumble in price as Nvidia's newer Blackwell generation ramped up.
Instead, surging inference demand from AI agents and coding tools absorbed capacity faster than it could be built, with clusters reportedly booked out until autumn.
Prices have since cooled from a May spike, which is why the Polymarket consensus clusters in the middle of the range.
If rates hold above $2.30, it suggests AI demand is still outrunning supply; a slide below $2.00 would be the first hard evidence the compute boom is easing.
