AI Giants Turn to Physical Books to Feed Data Hunger
AI-generated image · Bay Street Wire
Reports reveal Amazon and Anthropic are utilizing destructive scanning processes to acquire high-quality training data as digital sources dwindle.
Frontier AI developers are increasingly targeting physical books to secure high-quality training data that has not been generated by AI, BetaKit reports. As large language models exhaust available online text, companies are turning to bulk purchases of secondhand books.
According to BetaKit, 404 Media recently used a tracking device to confirm that Amazon has been buying books in bulk and scanning them at a Las Vegas warehouse, a process that involves cutting the spines and destroying the volumes. Amazon stated the company purchases these books to "help develop and improve the products and services our customers use."
BetaKit also notes that Anthropic led a similar initiative known as "Project Panama." Internal documents described this as an "effort to destructively scan all the books in the world" for AI training. The scale of these operations is significant; one vendor's proposal requested between 500,000 and two million books over six months, a volume that could encompass the entire 1.2 million-book physical catalogue of Montréal's Grande Bibliothèque.
This aggressive pursuit of data extends beyond books. BetaKit reports that Google and Mercor recently competed for access to legacy Spirit Airlines data, which included employee productivity records, Microsoft Teams chats, and hundreds of millions of internal emails. Google ultimately won the bid.