SellYourAIData
Get a free valuation Contact

AI Data Deal TrackerDeal brief

×

Harvard releases nearly 1 million public-domain books for AI training, funded by Microsoft and OpenAI

Harvard's Institutional Data Initiative published Institutional Books 1.0, a dataset of nearly one million public-domain books for AI training, supported by unrestricted gifts from Microsoft and OpenAI. The dataset is shared openly on Hugging Face.

CAMBRIDGE, Mass., June 12, 2025. Harvard's Institutional Data Initiative released Institutional Books 1.0, a dataset of nearly one million public-domain books for training AI models. The initiative is supported by unrestricted gifts from Microsoft and OpenAI.1

Terms of the deal

The dataset contains nearly one million books, some dating to the 15th century, in 254 languages. It includes more than 394 million scanned pages and an estimated 242 billion tokens. Fewer than half the volumes are in English, and the largest concentration of works is from the 19th century.1 It is shared on Hugging Face.1

Google worked with Harvard to retrieve public-domain volumes from Google Books, which began digitizing in 2006.1 The gifts from Microsoft and OpenAI are described as unrestricted, and no amounts are given. The licence terms of the dataset are not specified in the report.1

Background

This is a funding arrangement and an open release, not a sale. Microsoft and OpenAI support the initiative but did not buy exclusive rights. OpenAI also donated $50 million in 2025 to a group of research institutions, including Oxford's Bodleian Library, through its NextGenAI program.1

The initiative was first publicized in December 2024. Its stated goal is to level the playing field for researchers, startups and open-source developers, and the collection is roughly five times larger than Books3, the dataset used to train Meta's Llama.2

What it means for data sellers

Not every dataset needs a buyer to be valuable. Public-domain and institutional collections can attract sponsors who pay to have them cleaned and released. A company holding material with no copyright limits could consider funding or partnership models in which the value comes from curation, metadata and trust, not from exclusivity.

How the deal came together

  1. Harvard formed a data initiative. The Institutional Data Initiative was created to make curated institutional collections available for AI.
  2. Google helped retrieve the volumes. Google worked with Harvard to retrieve public-domain volumes from Google Books, which began digitizing in 2006.
  3. Microsoft and OpenAI funded the work. The initiative is supported by unrestricted gifts from both companies.
  4. The dataset was shared openly. It was posted on Hugging Face so anyone can download it.

Takeaways for companies with data

  • Public-domain collections can be prepared and released as a shared resource.
  • Funders gave unrestricted gifts, not purchases, so there is no exclusive access.
  • Curation and metadata are what turn old scans into a usable training set.

Sources

Every numbered reference above points to an entry here. Figures come from these pages as of the date shown at the top of the article. Where a source reports a number it got from unnamed people, the text says so.

  1. Harvard, Microsoft and OpenAI support a public-domain books dataset (Fortune, June 2025)
  2. Harvard Releases Massive Public-Domain Dataset for AI Training (Maginative, December 2024)

###

About the AI Data Deal Tracker

SellYourAIData tracks publicly reported sales and licenses of data to AI companies, with the price, terms and sources for each. Entries are checked against public reporting and updated as deals change. SellYourAIData was not involved in this deal. To ask what your company's data could sell for, get a free valuation. To report a deal, contact us.