An analysis of logistics chains and field data indicates that Amazon has launched a large-scale operation to digitize printed publications, which appears to be directly linked to training artificial intelligence algorithms. This is not just about scanning library collections, but about the industrial destruction of physical media, including rare and collectible copies.
The hidden route: from a used-book dealer to a secret workshop
My sources among used-book dealers have recorded an anomalous surge in orders: anonymous buyers are purchasing hundreds of books on a wide variety of topics and publication years. In one of these orders, placed through the Biblio marketplace, a tracker was discovered. The device led to an Amazon facility in North Las Vegas, where the VGT3 division is stationed.
An examination of open data on this division's activities and interviews with former employees allow us to reconstruct the full picture. Huge batches of books regularly arrive at the site. First, staff register each edition by scanning barcodes and ISBNs. Then comes the most barbaric stage: the bindings are cut off the books, and they are turned into stacks of individual sheets.
The death conveyor for books
Next, the pages are fed into high-speed scanners. This approach does speed up the digitization process, but it makes physical restoration of the publication absolutely impossible. According to workers, scanning is the main and only task of VGT3. They describe a routine process: receiving, accounting, cutting, and digitization.
Notably, in early 2026, the facility experienced disruptions in the supply of raw materials—books were in short supply—but the flow soon recovered. This indicates that the program is not a one-off but a systematic effort.
In response to inquiries, the corporation confirmed that it does indeed purchase printed publications through commercial channels but refused to disclose volumes, timelines, and target products. The official wording—"development and improvement of products and services"—leaves no doubt that this is about training large language models.
My analysis: This situation exposes a troubling trend. We are witnessing a paradox: to create "intelligence" that is supposed to preserve knowledge, its physical carriers are being destroyed. While the industry pays for data without considering cultural value, we risk losing a significant portion of the book heritage. This is not just a business issue—it is a matter of ethics and the preservation of history.