AI Infrastructure Index

The Stockatomics AI Infrastructure Index follows what the hardware behind AI costs, month by month. It brings together the chips, memory and optics inside data centres with the transformers, switchgear and cooling that power them, and reads 100 in January 2023.

January 2023 = 100, latest month August 2026

Open in ChatGPTOpen in ClaudeOpen in Perplexity

AI Infrastructure Index

Monthly, January 2023 = 100

02505007501,0001,2501002023202420252026

AI Infrastructure IndexJanuary 2023 = 100

What the index tracks

An AI data centre is a building full of hardware. Accelerators and memory do the computing, optical transceivers move data between them, and circuit boards and capacitors hold it all together. Around them sit the switchgear, transformers, converters, turbines and cooling systems that keep everything powered and running.

The AI Infrastructure Index follows what US buyers pay for each unit of that hardware when it’s imported, and combines it into one number. When the index rises, the equipment going into AI infrastructure costs more per unit, and when it falls, it costs less.

The index starts at 100 in January 2023. A reading of 200 means the hardware costs twice as much per unit as it did then, while a reading of 50 means it costs half as much.

Some of the rise comes from prices climbing as demand outruns supply. Some comes from what’s being bought, as data centres move from standard servers to far more powerful and expensive AI systems. Together they show how much more capital each piece of AI infrastructure now takes.

Every month, each component’s Stockatomics benchmark, the average value of a unit imported into the US, is compared with its level in January 2023. The index combines those changes into a single figure, giving each component a weight based on how much the US imported of it in 2025, so the components with the most money flowing into them count the most.

The index updates every month, as new benchmarks are published.

A monthly measure of what the hardware behind AI costs, published by Stockatomics. It follows the chips, memory and network optics inside data centres, as well as the power and cooling equipment around them, and reads 100 in January 2023.

It shows how much the cost per unit of AI hardware has changed since January 2023. A reading of 300, for example, means that hardware costs three times as much per unit as it did then.

Two things push it up. Prices have climbed as demand for AI hardware outpaced supply, and each unit bought has become more valuable as data centres switched from standard servers to far more powerful AI systems. The index captures both.

The hardware that AI data centres are built from, from accelerators, memory and optics to transformers, switchgear and cooling. The full list, with how much each one counts, is on this page.

By how much the US imported of each one in 2025, so the components with the most money flowing into them count the most.

No. It’s a level that compares each month with January 2023, which makes components priced per chip, per transformer or per kilogram comparable. Each component’s own price in US dollars is on its page.

No. It tracks the cost of physical hardware, not share prices. To see how the companies that make and buy this hardware are performing, open any component’s page.

Every month, when the latest benchmarks are published.

On this page and in the Stockatomics Terminal.