An investigative report by technology outlet 404 Media has uncovered an unusual data collection practice by Amazon. The company is acquiring large volumes of old, out-of-print, and rare physical books. The primary goal of these acquisitions is to extract high-quality text data to train proprietary artificial intelligence models.
To verify the operational pipeline, journalists placed a tracking device inside a package of rare books purchased from an independent seller. The tracked shipment traveled directly to an Amazon facility in Las Vegas, Nevada, operating under the facility code VGT3. Worker accounts and internal reports confirmed the activities taking place inside the building.
Upon arrival at the Las Vegas center, the volumes are prepared for automated scanning rather than storage or retail distribution. Workers cut off the spines of the books so that loose pages can be fed rapidly into industrial high-speed scanners. This procedure permanently destroys the physical copies in exchange for high-resolution digital text captures.
The aggressive digitization effort highlights a broader challenge across the machine learning industry. AI developers are increasingly struggling to find clean, copyrighted, and human-written datasets that are completely free from synthetic AI slop. Physical books published decades ago provide a guaranteed source of human-authored language.
When questioned about the findings, Amazon confirmed that it purchases physical books through commercial channels. A company spokesperson stated that these materials are utilized to develop and refine various internal products and services.