Wikipedia has announced AI content licensing deals with major technology companies, including Microsoft, Meta, Perplexity, and Mistral AI. The Wikimedia Foundation announced this on January 15, 2026, as part of its 25th anniversary celebrations. These agreements expand the reach of Wikimedia Enterprise, the Foundation’s commercial platform that provides structured, high-volume access to Wikipedia’s vast corpus of human-curated content for AI training and other advanced uses.
According to the Foundation, the Enterprise programme enables companies to access content “at a volume and speed designed specifically for their needs,” offering structured data access that goes beyond informal scraping of the publicly accessible site, while supporting the nonprofit’s mission to keep Wikipedia free and open for all.
Furthermore, Lane Becker, senior director of Wikimedia Enterprise, explained to The Verge, “We take feature requests, we build features and functionality, and sort of try to structure the data in ways that support these companies’ needs.” She added that “it is in every AI company’s best interest to support the long-term sustainability of Wikipedia, because Wikipedia and all the other projects that we support are so core to their business.”
How Does Wikimedia Enterprise Work?
The Wikimedia Foundation created Wikimedia Enterprise as a commercial product to make large-scale reuse of Wikipedia and other Wikimedia project content more efficient, reliable, and suitable for high-volume use. It was launched on October 25, 2021, to meet the needs of companies and organisations that require frequent, structured access to Wikimedia’s data.
Unlike the publicly available APIs and data dumps that serve casual users and developers, Wikimedia Enterprise delivers enterprise-grade APIs (Advanced APIs designed for large-scale commercial use) that provide machine-readable, parsed, and structured content at scale. This ensures that partners can quickly access and process information across Wikipedia and other Wikimedia projects without investing in extensive internal tooling to scrape or clean raw data.
Moreover, Wikimedia Enterprise operates as an opt-in commercial service backed by enterprise contracts and service-level agreements, offering companies a more reliable alternative to public APIs and data dumps. The Wikimedia Foundation said the platform was built specifically for organisations that require high-volume, commercial reuse of Wikimedia content and need efficient access supported by formal agreements.
Importantly, most of the data remains available under free Creative Commons licences, with metadata clearly explaining the licence terms. Wikimedia Enterprise adds value through enhanced access, advanced consulting, and data management, rather than replacing free access.
How AI Bots Drive Up Wikipedia Costs
The Wikimedia Foundation has warned that automated scraping by AI bots and crawlers is driving up costs due to increased bandwidth demand, a burden that its donation-funded model struggles to absorb. According to the Foundation, bandwidth used for downloading multimedia content has surged by roughly 50% since January 2024. This rise is not coming from human readers but primarily from automated programmes collecting data for AI training.
Much of this traffic consists of bots scanning Wikimedia Commons’ repository of 144 million images, videos, and other media, which is more expensive to serve from core data centres than popular cached pages.
Furthermore, Wikimedia found that bots account for about 65% of the most costly requests to its infrastructure, despite representing only around 35% of total pageviews. This is because bots tend to request less-popular content that falls outside regional caches and must be served directly from core data centres. This pattern places disproportionate demands on computing resources and contributes to rising operational expenses.
At the same time, Wikimedia’s own user trends update showed that real human pageviews declined by around 8% year-on-year in 2025, in part because search engines and generative AI tools trained on Wikipedia content increasingly answer questions directly without directing users back to Wikipedia pages.
Overall, this combination of rising automated load and declining reader traffic reflects a structural cost pressure on Wikipedia’s infrastructure in the AI era, with co-founder Jimmy Wales acknowledging that non-human traffic is financially taxing due to higher server costs.
Why This Matters
Wikipedia sits at the centre of the internet’s knowledge economy. It underpins search engines, voice assistants, knowledge graphs, and now generative AI systems used by hundreds of millions of people every day, while also serving as a free source of knowledge for netizens worldwide. However, as AI developers and data companies increasingly rely on Wikipedia’s content at an industrial scale, the nonprofit’s open model is coming under unprecedented technical and financial strain.
On the one hand, Wikipedia’s articles, images, and structured data have become core training material for large language models and AI search tools. As a result, automated bots now generate a large share of the platform’s most expensive traffic, driving up bandwidth and server costs while contributing little to the donation system that funds the site.
At the same time, human pageviews are falling as AI tools answer questions directly, reducing the visibility of Wikipedia itself and weakening a key source of donor engagement.
This shift matters because Wikipedia operates on a razor-thin margin. The Wikimedia Foundation runs one of the world’s most visited websites while remaining donation-funded and free to use. Yet it now faces a future in which machines, not people, consume much of its content, and do so at a scale that strains infrastructure built for human readers.
In effect, Wikipedia has become a critical piece of digital infrastructure for the AI era. How it balances open access with sustainability will shape not only the future of free knowledge, but also the economics of the wider AI ecosystem.
Curated by James Chen
