Last week, an employee at Amazon’s VGT3 warehouse in Las Vegas described the process of stripping spines from thousands of books to feed them into scanning machines for artificial intelligence training.
The warehouse floor
The facility sits alongside LAS8, Amazon’s print-on-demand operation. Workers confirmed that the site receives shipments containing rare volumes, some sourced from the University of London and others liquidated from libraries.
One employee noted that staff often admit they do not fully understand the purpose of the scanners. The workflow moves quickly. Staff unbox arrivals and stack them into plastic totes. Pallet jacks then transport these stacks to a receiving area where barcodes are scanned to identify duplicates.
Items deemed unnecessary are placed into tall cardboard shuttles, described as six or seven feet high. These bins are then moved directly to the cutting station. The employee stated that once a book enters this stage, it is impossible to reassemble.
The destruction process
Each cutting station is operated by a single person. A machine features a protective cover and a sliding slot for the volume. After the operator removes their hands, a blade descends to slice through the spine. The pages are then separated and fed into high-speed scanners that resemble cash-counting devices.
On the screens, the text flips rapidly. The scanned data is used to train AI models, a purpose the employee confirmed after initially suspecting the operation was for Kindle devices.
What the worker thinks
The staff member expressed concern over the loss of physical copies. They noted that some of the incoming material includes government documents presented to Parliament on behalf of the Queen. The employee argued that destroying these items reduces the amount of available knowledge for other readers.
What it means
For authors and publishers, this confirms that physical inventory is being converted into digital training data without the traditional licensing models used for digital publishing. The sheer volume of material, including sealed new books and obscure archives, suggests a shift in how training data is sourced.
