Amazon VGT3 cuts bindings from rare books at Las Vegas site for Nova AI scans

Amazon acquired bulk shipments of rare and foreign-language books, tracked via a 404 Media device to the VGT3 warehouse. Employees there process volumes by guillotine-cutting spines for high-speed flatbed scanning, after which the originals are discarded. This method matches the industrial approach Google Books used after 2004 but exceeds it in scale for current large language model pipelines.

Internet Archive documentation states its operators digitize one page at a time without spine removal to preserve both content and artifact integrity. That workflow requires custom fixtures and longer cycle times per volume. Primary source logs from the Archive show cumulative output exceeding 5 million preserved titles since 1996 with zero reported binding destruction events.

Fair use litigation records, including the 2023 Authors Guild filings and the ongoing New York Times v. OpenAI docket, treat text extraction as reproduction but leave physical destruction unaddressed. Amazon’s choice to destroy rather than loan or return copies therefore sits outside existing precedent yet aligns with the same training-data acquisition pattern used by multiple frontier labs.

Operational outcome: Amazon can now claim possession of tokenized text from low-circulation works that no public archive holds in digital form, while incurring no ongoing storage cost for the originals.