AI compute market signals and learning

The market terminal for AI compute.

Daily benchmarks, analysis, and intelligence on the AI compute market — so you can build, buy, and invest with confidence.

  • Fact-based Benchmarks
  • Plain-English Analysis
  • Updated Daily
  • Free & Independent

What the benchmarks say

Updated July 21, 2026 at 6:00 AM EDT
See all benchmarks

Term of the Day

Benchmark

A benchmark is a consistent reference point used to compare prices, costs, or market conditions over time. ComputeTape uses benchmarks to make AI compute markets easier to read. A good benchmark should be clear, source-backed, repeatable, and useful for decision-making.

Learn more

Today's Lesson

How to Estimate Monthly AI Compute Burn

Monthly AI compute burn measures recurring spending on the capacity used to train, fine-tune, experiment, and serve models.

Start the lesson

Providers

On-demand GPU prices from tracked providers

Current reviewed list price per GPU-hour and the H100-equivalent rate for each provider, each from a public price page with an editorial check and how long that price has held. Rows outside the review window are kept in the source data but not shown.

  • AWS$6.88/GPU-hr
    HyperscalersH100us-east-1
    Held 30 days
  • Azure$12.29/GPU-hr
    HyperscalersH100eastus
    Held 66 days
  • CoreWeave$6.16/GPU-hr
    AI-native cloudsH100us
    Held 3 days
  • Crusoe$3.90/GPU-hr
    AI-native cloudsH100Multi-region
    Held 54 days

Explore ComputeTape

Compute College

Learn the AI compute market, one lesson at a time.

66 plain-English lessons across 8 tracks — from GPUs and model costs to power, data centers, and market structure. Each one answers a real question, then shows how it moves compute cost, supply, and pricing.

Explore Compute College