Microsoft, Meta, Amazon, Perplexity, and Mistral AI are now paying Wikipedia for structured access to its 65 million articles through the Wikimedia Enterprise platform. The Wikimedia Foundation announced the partnerships on January 15, 2026, timed to Wikipedia's 25th birthday.
These companies join Google, which signed on back in 2022, along with smaller partners like Ecosia, Nomic, and ProRata.
The scraping problem
The deals address a growing financial strain. According to Wikimedia's own data, bandwidth used for downloading multimedia content has surged roughly 50% since January 2024, driven almost entirely by AI bots rather than human readers.
More telling: bots account for about 65% of the foundation's most expensive server traffic. That's because automated scrapers tend to pull obscure, less-cached pages that require routing through core datacenters rather than regional caches. The foundation discovered some AI companies were even disguising their crawlers with rotating IP addresses and spoofed user-agent strings to evade detection.
Human pageviews, meanwhile, dropped around 8% year-over-year. The irony is not lost on anyone: AI systems trained on Wikipedia content are now answering questions that used to send people to Wikipedia.
What the enterprise product offers
Wikimedia Enterprise provides three API tiers. The On-demand API returns current article versions on request. The Snapshot API delivers Wikipedia as a downloadable file updated hourly. The Realtime API streams edits as they happen. All three offer cleaner, structured data than scraping the public site.
Lane Becker, president of Wikimedia Enterprise, told Reuters that the foundation spent time refining the product to meet corporate needs. The challenge was convincing companies to pay for something they had been harvesting for free.
Microsoft's corporate VP Tim Frank framed the partnership around supporting reliable information online, calling it part of building a "sustainable content ecosystem for the AI internet."
The volunteer question
Wikipedia's 250,000 volunteer editors continue to write, fact-check, and maintain all the content. None of them get paid, even as their work trains billion-dollar AI products. The enterprise deals create some financial return to the foundation's infrastructure, but the gap between volunteer labor and commercial extraction remains awkward.
Perplexity, for its part, offered 2,500 Enterprise seats to share with Wikipedia editors as a birthday gift, according to a PYMNTS report. A nice gesture, though it does nothing to address the fundamental asymmetry.
Leadership transition
The announcement coincides with broader changes at Wikimedia. Bernadette Meehan, a former U.S. Ambassador to Chile and Obama Foundation executive, takes over as CEO on January 20, succeeding Maryana Iskander. The foundation announced her appointment in December 2025.
Meehan inherits a platform caught between its open-access mission and the commercial realities of the AI era. Wikipedia's terms allow anyone to reuse content under Creative Commons licensing, but hosting that content for industrial-scale extraction costs real money that donations barely cover.
Whether these enterprise deals solve the sustainability problem or just buy time remains unclear. The foundation's 2025/2026 annual plan sets a goal to reduce scraper traffic by 20% in request rate and 30% in bandwidth. Paid API adoption might help, but plenty of AI companies may still prefer the free approach.
