As generative AI companies search for cleaner training data, one of the internet's oldest institutions is quietly changing its economic model. The Wikimedia Foundation, which operates Wikipedia, has confirmed new agreements with major AI players, including Amazon, Meta, Microsoft, Mistral AI, and Perplexity. The deals formalize paid access to the encyclopedia's vast information trove – content that has long functioned as both an open resource and a magnet for automated web scrapers.
The foundation said the contracts give participating companies access to structured Wikipedia data at scales and speeds tailored to their requirements. The organization did not disclose financial terms. Even so, the move marks a turning point for one of the world's most visited websites, shifting from a model mainly built on small donations toward commercial partnerships with companies developing the next generation of large language models.
Foundation executives say this strategy is a response to soaring technical demands on the network. Automated scraping – often disguised as regular traffic – has intensified as AI developers harvest online text for model training. As a result, the load on Wikipedia's servers has grown significantly, even as human readership has fallen by roughly eight percent over the past year.
Wikimedia operates one of the internet's most complex server ecosystems, hosting more than 65 million articles across roughly 300 languages, edited by about 250,000 volunteers. Wikimedia Foundation Chief Executive Officer Maryana Iskander told The Associated Press that maintaining the data infrastructure supporting both human readers and machine access comes at a significant cost.
"Our infrastructure is not free," Iskander said. "It costs money to maintain servers and other infrastructure that allows both individuals and tech companies to draw data from Wikipedia."
Wikipedia founder Jimmy Wales has welcomed the partnerships as a practical solution. He argued that models trained on Wikipedia benefit from its human editing process, which filters out misinformation and enforces verification standards.
"I'm happy that AI models are training on Wikipedia because it's human-curated," he said, adding that "[AI firms] should chip in and pay for their share of the cost that [they're] putting on us."
The debate over data reuse has been contentious across the tech industry. While image libraries and publishers have pursued legal action against unauthorized use of data for training, Wikimedia has taken a different path. Rather than restrict access, the foundation is steering toward collaboration and compensation, acknowledging how Wikipedia's open structure has made it central to the AI ecosystem – and how sustaining that openness requires funding.
At the same time, Wikimedia is exploring its own uses for artificial intelligence. Wales described plans to develop tools to automate routine editorial maintenance, such as identifying broken links and recommending source replacements based on contextual analysis. These systems wouldn't replace human editors, he said, but could reduce repetitive work. He also envisioned a future in which Wikipedia's search evolves into a conversational engine that can quote directly from verified text in response to user queries.
Wikipedia's journey spans 25 years of collaborative publishing, controversy, and adaptation. The platform remains one of the internet's top ten destinations and a frequent flashpoint in cultural and political debates.
Critics, including some US lawmakers and tech figures such as Elon Musk, have accused Wikipedia of ideological bias – a charge Wales dismisses as inevitable in polarized online discourse. Musk's own AI-driven rival, Grokipedia, mirrors Wikipedia's format but relies on large language models that, according to Wales, cannot yet match the encyclopedia's accuracy or editorial depth.
Despite the turbulence, Wikimedia's leadership frames the latest deals as a pragmatic recalibration rather than a retreat from its founding ideals. The nonprofit still draws most of its revenue from roughly eight million individual donors. However, enterprise customers now provide a new source of capital in an era where the largest consumers of its data are machines, not people.
Wikipedia is now getting paid by Meta, Microsoft, Perplexity, and other AI companies
