AI isn’t going anywhere. If anything, it’s becoming more deeply woven into how people search, learn, and get answers online. But for AI systems to actually get better, more accurate, less hallucination-prone, and more useful, they still need something they can’t generate on their own: large volumes of unfiltered, human-curated knowledge.
For years, that need has put AI companies on a collision course with creators and platforms whose content was being scraped at scale, often without permission or compensation. Now, that tension is starting to ease, and Wikipedia may be showing what a more collaborative future looks like.
Ogbonda Chivumnovu
The Wikimedia Foundation, the nonprofit behind Wikipedia, has signed new commercial data-access deals with several major AI and tech companies, including Microsoft, Amazon, Meta, Perplexity, and France’s Mistral AI. The agreements fall under Wikimedia Enterprise, a paid product that allows companies to access Wikipedia’s data in a structured, high-volume format designed specifically for large-scale use like AI training. While some of these partnerships were finalized over the past year, Wikimedia only publicly confirmed them as part of its 25th anniversary.
This is a significant shift from the free-for-all that defined early AI training. Wikipedia’s more than 65 million articles across 300 languages are among the most valuable datasets for training large language models. But as AI companies increasingly relied on automated scraping, Wikimedia saw server costs rise while human traffic declined.
In 2025, the foundation reported an 8% drop in pageviews compared to the same period in 2024, partly because generative AI tools were answering questions directly without sending users back to the site. The result was a growing imbalance: Wikipedia was powering AI products while relying mostly on small public donations to stay online.
Rather than blocking AI companies or heading to court, a route some publishers have taken, Wikimedia chose to monetise access. Under the new model, AI companies pay to use Wikipedia’s content at the scale and speed their systems require, while Wikimedia gains funding to support its infrastructure and volunteer-driven editing community. But it's still not clear just how much Wikipedia is getting. For AI developers, the benefit is clear: reliable, human-governed data that improves model quality without legal uncertainty or unstable scraping methods.
For Wikipedia, the move helps protect one of the last major free-knowledge projects of the early internet, without abandoning its open ethos. The foundation has stressed that it isn’t anti-AI. It wants AI to work, but in a way that’s sustainable. As AI continues to reshape how information is consumed, these deals suggest the industry may finally be moving from extraction to cooperation, where the human knowledge powering AI is treated as something worth paying for.
Emmanuel Umahi
