Wikipedia Now Charges the AI Giants It Fed for Free

AI–generated answers drained 8% of its human traffic and a Musk-built clone is copying its pages, so the nonprofit behind Wikipedia is turning its 65 million articles into a paid product for Amazon, Google, Meta, and Microsoft.

The Wikimedia Foundation, which runs Wikipedia, now charges Amazon, Google, Meta, Microsoft, and other AI firms for bulk data access through its commercial arm, Wikimedia Enterprise. That revenue jumped 148% to $8.3 million in the last fiscal year. The move comes as AI-generated summaries cut Wikipedia’s human page views by 8% and quietly threaten the donation engine that actually funds the site.

For 25 years, Wikipedia’s deal with the internet was simple. Take the knowledge, it is free. Artificial intelligence just made that deal untenable. The companies training the world’s chatbots were downloading the encyclopedia wholesale, then serving its facts back to users who never clicked through. So Wikipedia did something it had long resisted. It put a price on the door.

What Happened

The Wikimedia Foundation confirmed a roster of paying AI customers for Wikimedia Enterprise, its four-year-old commercial product that sells fast, structured, high-volume access to Wikipedia’s content. Google became the first major customer back in 2022. Now Amazon, Meta, Microsoft, France’s Mistral AI, and Perplexity have joined, alongside smaller players like Ecosia, Nomic, and ProRata.

The reason is a shift in who actually reads Wikipedia. Human page views of the English edition fell 8% year over year, as AI Overviews in search results and chatbot answers hand people Wikipedia’s facts without sending them to Wikipedia. At the same time, automated scrapers now account for close to a third of the site’s most demanding bandwidth, and the foundation can currently block only about 30% of the roughly 1.5 billion abusive data requests it sees each day.

New chief executive Bernadette Meehan, a former US ambassador to Chile who took the role on January 20, 2026, is running a two-track response. Charge the heavy commercial users, and throttle the freeloaders.

The Backstory

Wikipedia launched in 2001 and became one of the ten most visited sites on the internet on a model that looked impossible on paper. Roughly 250,000 unpaid volunteers write and fact-check more than 65 million articles across 300-plus languages, funded almost entirely by small reader donations. It is the textbook case of the crowdsourcing business model, where the crowd supplies the product and the organization only has to keep the platform running.

That model built the single cleanest large dataset on the open web, which is exactly why AI developers wanted it. Wikipedia is a core training source for the chatbots now competing for the same attention. Wikimedia launched Enterprise in January 2022 to charge the biggest reusers, but for two years the money was trivial: $3.2 million in fiscal 2022-23, then $3.4 million in 2023-24. The AI boom is what finally moved the number.

The Plan

Meehan’s job is to diversify revenue without breaking the thing that makes Wikipedia trusted: its independence. Enterprise is the commercial lever, but the foundation has capped it at 30% of total revenue, a rule formalized in May 2025, so the site never becomes beholden to a handful of tech clients.

Around that, the plan is defensive breadth. Wikimedia nearly doubled its lobbying spend to $174,300 and retained the law firm Holland & Knight at $150,000 a quarter. It bought Times Square advertising, is chasing younger contributors through TikTok and Roblox, and expanded a human rights team that now shields volunteer editors from doxxing, surveillance, and arrest.

The Business Model Angle

Here is where the easy headline breaks down. “Wikipedia is now an AI data business” is the story everyone wants to tell. The numbers say otherwise, and the real read is sharper.

That $8.3 million is only 4% of the foundation’s $208.6 million in total revenue. Donations still supply $189.5 million, more than 90% of the money. Enterprise is a hedge, not a rescue. What makes Wikipedia’s data worth paying for is the one thing AI cannot cheaply reproduce: hundreds of thousands of humans arguing their way to a verifiable, neutral record. For 25 years that asset was given away. AI turned it into something with a market price, and Wikipedia is finally collecting.

The trap is on the other side of the ledger. The banner and email donations that fund the operation depend on human beings landing on Wikipedia pages. AI answers are cutting exactly that traffic. So the same Google and Microsoft products now paying for Enterprise access are also draining the reader funnel that supplies the donations. Wikipedia is selling premium access to the companies disrupting its core revenue. The license income has to grow faster than the donation base erodes. That race, not the $8.3 million, is the whole business.

The Risk

The clearest risk is that the hedge is too small to matter in time. If AI keeps eroding human traffic, an 8% annual decline compounds into a serious donation problem long before Enterprise scales to fill the gap.

The second risk is competitive and personal. Elon Musk has attacked Wikipedia as biased, urged followers to withhold donations, and built Grokipedia through his xAI company, a clone that lifts liberally from Wikipedia, keeps no public edit history, and invents its own errors. A customer ecosystem that also spawns direct competitors is an unstable one.

The third is political and legal. A Trump administration official has probed the foundation’s tax-exempt status, and a Republican-led congressional committee opened an investigation demanding editor identities. Globally, governments filed 117 content requests last year, up from 10 a decade ago, most from Russia. Defending volunteers and projects now consumes 32% of the budget, and litigation costs have doubled since 2020.

Quick Questions

The Business Model Analyst Take

Wikipedia spent 25 years proving the best data on the internet could be given away. AI is forcing it to learn the opposite lesson: when your content becomes the raw material for a trillion-dollar industry, “free” is a subsidy you are paying to your own competitors. Charging Enterprise customers is plainly the right move, and capping it at 30% to protect independence is the disciplined version of it.

But do not confuse the licensing win for the main event. The number that decides Wikipedia’s future is not the $8.3 million it now earns from AI. It is the 8% of human readers it has already lost to AI, because those readers are the donors. Every content business watching this, including this one, should read the sequence plainly. When an AI layer sits between you and your audience, you either turn your data into a licensed asset or you watch the traffic that funded you get absorbed, summarized, and resold. Wikipedia has the rarest asset in that fight, a dataset no model can fake, and it still may not be enough. That is how hard this shift is.

Reporting drawn from The New York Times, the Associated Press, and the Wikimedia Foundation’s FY2024-25 audited financial reports.