Understand the cost and availability of AI compute.

Free Morning Brief: Wednesday, July 1, 2026

ComputeTape tracks GPU pricing, model costs, provider capacity, power constraints, and the infrastructure shaping the AI market.

Learn how GPUs, model costs, power, and data centers shape the market — in plain English.
Estimate GPU-hour, training-run, and model serving costs with free calculators.
Track daily pricing, capacity, and power signals in the Morning Brief.

Read today’s Morning Brief

Daily at 6:00 AM

Lesson of the Day

What is High-Bandwidth Memory (HBM)?

Learn the AI compute market one concept at a time.

High-bandwidth memory is fast memory located near advanced accelerators to keep AI workloads supplied with data.

Power & Data Centers

Read today’s lesson Explore Compute College

Compute College

Choose a learning path

Learn AI compute one short lesson at a time. Your progress is saved in this browser — no account required.

Track 1 · AI Compute 101Start: What is AI compute
0/7
Track 2 · GPU Pricing & CapacityStart: H100 price per hour
0/8
Track 3 · Model CostsStart: What is model training cost
0/7
Track 4 · Model Benchmarks & AI Compute EconomicsStart: What are AI model benchmarks
0/23
Track 5 · Power & Data CentersStart: Why power matters
0/17
Track 6 · Compute Market StructureStart: What is a neocloud
0/8
Track 7 · Compute FuturesStart: What are compute futures
0/5
Track 8 · Buyers & OperatorsStart: How AI labs buy compute
0/8

Open Compute College →

Learn the Market

Term of the Day: Power usage effectiveness

Power usage effectiveness, or PUE, compares total facility power to the power used by IT equipment. A lower PUE means less extra power is being spent on cooling, lighting, and other overhead. AI data centers care about PUE because GPU clusters already consume large amounts of electricity.

What is ComputeTape?

Your practical guide to the AI compute market.

ComputeTape helps you understand the cost and availability of AI compute — from GPU pricing and cloud capacity to power constraints, model training and serving costs, and the providers behind them.

Learn how the market works, estimate real compute costs, compare providers, and track the signals that show when AI compute is tightening, loosening, or shifting before it reaches the headlines.

Use ComputeTape to

Learn AI compute in plain English.
Estimate GPU-hour, training, and model serving costs.
Compare source-backed GPU providers and current GPU pricing.
Track daily price, capacity, and power signals.

Providers

On-demand GPU prices from tracked providers

Live 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 whose source has not been checked yet are kept in the data but not shown.

AWS$6.88/GPU-hr
HyperscalersH100us-east-1$12.29 H100eno change in 10 daysSource
Azure$12.29/GPU-hr
HyperscalersH100eastus$12.29 H100eno change in 46 daysSource
CoreWeave$8.60/GPU-hr
AI-native cloudsB200us$4.30 H100eno change in 21 daysSource
CoreWeave$6.16/GPU-hr
AI-native cloudsH100us$6.16 H100eno change in 46 daysSource