Technology / Economy

Add as a preferred source on Google

H100 Rental Measures Diverge, With Transactions at $2.74 an Hour

The advertised-rate sample had a preliminary median of $3.53 per GPU-hour for Oct. 12-18, based on 36 listings from 34 providers.

Updated 1 min read
Gloved hands position an H100 accelerator in a hardware validation fixture in Santa Clara, California. / EconoTimes.ai
Gloved hands position an H100 accelerator in a hardware validation fixture in Santa Clara, California. / EconoTimes.ai

Read the original article

Two measures of NVIDIA H100 rental costs gave different mid-October price signals for AI computing: a neocloud transaction index stood at $2.74 per GPU-hour, while advertised rates showed a provisional median of $3.53.

At 4 p.m. ET (20:00 UTC) Oct. 16, the completed-rental gauge for H100 SXM neocloud capacity read $2.74 per GPU-hour. It lost 1.8% during the prior week but remained 4.6% higher than its level 30 days earlier; its methodology gives more influence to higher-volume rentals and adjusts unusually large deviations.

An advertised on-demand price series placed its preliminary midpoint at $3.53 per H100 GPU-hour across Oct. 12-18, using 36 listed rates from 34 providers logged at 11:40 p.m. ET Oct. 16 (3:40 a.m. UTC Oct. 17), before the week ended.

The measures rely on distinct data and observation windows, so their amounts cannot be treated as a direct comparison. Both describe hourly rental access to H100 computing; they say nothing about secondhand hardware prices.

Related articles

View all