To Train Its AI, Amazon Is Snapping Up Rare Books and Cutting Off Their Bindings

How Amazon Is Building a Rare-Book Collection for AI

According to НВ — Техно: It came to light at 12:30 on August 18 that Amazon has been buying up uncommon books and slicing off their spines so the pages can be scanned for AI training data. The news came from 404 Media, whose journalists attached a tracker to one such rare volume and followed its journey. That book ended up at an Amazon facility in Las Vegas known as VGT3, where the site features a dinosaur depicted holding a book in its claws.

The company is purchasing large quantities of titles that are out of print or otherwise difficult to find in digital form. After acquisition, these books may be broken apart for scanning. In a statement to 404 Media, Amazon confirmed that it buys books through commercial channels, saying the practice helps improve the products and services customers use.

Why Rare Books Matter for Language Models

Developers of large language models have a strong reason to seek out rare books: these systems need massive collections of text for training. Much of the content already freely available online has been used for that purpose, which makes physical-only publications especially valuable. Older texts that appeared before today's generative models were created were far more likely to be written by humans. This is crucial because a model trained on artificial intelligence output can suffer from model collapse, leading to lower-quality responses. For tech companies, old and scarce books offer a source of original written material.

Amazon's practice points to a broader tech-industry hunger for rare and unjustly overlooked texts, which may unlock new paths for artificial intelligence development.

404 Media

Now that most accessible online information has already been consumed, scarce publications could become a key reservoir for boosting AI training quality. The strategy also raises questions about cultural heritage preservation and intellectual property, since libraries and archives may face pressure from commercial interest in these materials.