Storage price trends - live cost per terabyte history

Why a price tracker keeps its history

Most price comparison sites show you today and throw yesterday away. That makes them useful for one question, what is cheapest right now, and useless for the question people actually ask during a shortage, which is whether the number in front of them is normal.

A price is only meaningful against a baseline. $18 per terabyte means nothing on its own; it means a great deal if the category has been sitting at $26, and it means the market has moved past you if it has been at $14 all quarter. Everything on these pages exists to supply that baseline from observations rather than from memory or vibes.

The honest limitation is stated on every page: we started keeping snapshots on a specific date, and comparisons that need more history than that print a dash. There is a strong temptation to stretch a few months into a year-over-year headline, and that temptation is exactly how confident-sounding wrong numbers enter circulation. The window grows on its own. Until it does, the pages say what they can support.

Everything here is free to cite, screenshot and embed with a link. If you need a cut we do not publish, ask.

Methodology

• Source. Our own tracking of Amazon US listings through the Amazon Creators API. One retailer, stated plainly, rather than an unnamed blend.
• Cadence. Prices refresh every few hours. One snapshot per listing per day is written to the price history, and these charts plot those daily snapshots.
• Window. Tracking for this category begins 2026-03-23 and runs to 2026-08-07. Comparisons longer than that window print a dash rather than a number.
• Cost per tb. Listed item price divided by the manufacturer's stated capacity. Shipping and tax are not included. Server RAM is quoted per gigabyte; storage per terabyte.
• Levels. The cheapest and median figures are plain observations across every listing we priced that day in that category and condition. They tell you where the market sits.
• Changes. Every percentage change is a chained matched-model (Jevons) index, not a comparison of two days' medians. Each daily step compares only listings we priced on both that day and the day before, taking the geometric mean of their individual price relatives, so a product entering or leaving the catalogue cannot move it. Comparing medians across dates would measure catalogue churn as if it were price movement, and on this dataset that difference has been large enough to invert the story.
• Condition. New and recertified are separate series throughout. Refurbished enterprise stock is frequently the cheapest per terabyte, and blending it into one line would hide both that fact and any change in its supply.
• Coverage. 498 priced listings in this category on the most recent day. Out-of-stock listings keep their last observed price in the history and are labelled where they appear in tables.
• What we do not do. No modelled projections, no smoothing, no imputed values for days we did not observe. A gap in a line is a gap in the data.

Citing this page

Quote a figure, link to the page, and tell readers where the number came from. That is ordinary citation and it needs no permission from us.

We do not offer this as an open dataset, and we removed an earlier notice that did. The underlying prices are retrieved under a licence that does not permit us to pass redistribution rights on to anyone else, so bulk reuse, republication of the series, or building a product on top of these numbers is something to ask about first rather than something we can grant in a footer. The contact page reaches us, and we would rather have the conversation than have you rely on a permission we cannot give.