Every day, ordinary people open a familiar website and type something completely unremarkable: a question about a car repair, a hot take on a TV show, a recipe tip left in a comment thread at midnight.
Nobody thinks twice about it.
But behind those throwaway posts, a market has been quietly taking shape, one that most users have never heard of, and that is now producing numbers that are genuinely hard to believe.
A market that barely existed three years ago is now worth billions
The global AI training dataset market size was valued at $3.59 billion in 2025 and is projected to grow from $4.44 billion in 2026 to $23.18 billion by 2034.
That growth is being driven by a single, urgent problem that every major AI company is quietly losing sleep over.
The models that power today’s most advanced AI systems were trained on enormous amounts of human-written text, and that text came largely from the open internet.
The question now is what happens when that supply runs out.
The internet writes fast, but the AI models are reading even faster
Epoch AI, a research organization that tracks AI scaling constraints, projects that the stock of publicly available human-generated data will be fully utilized between 2026 and 2032.
Epoch AI estimates the total effective stock of human-generated public text data at around 300 trillion tokens, counting only data that is sufficiently high-quality to be used for training and accounting for the possibility of training models for multiple epochs, but frontier models consume data faster than the internet produces it.
That creates a scarcity problem that no one in Silicon Valley has a clean answer to.
It also means that existing human-written archives, the kind sitting on popular websites right now, have become extraordinarily valuable almost overnight.
This is where the overlooked website enters the picture.
One platform quietly became the most contested piece of real estate in AI
Think about what a major social platform really is, from an AI company’s perspective.
It is decades of real human conversation: debate, humor, advice, opinion and lived experience, all written in natural language, all organized by topic, and all time-stamped.
For a model trying to learn how people actually think and write, that is not just useful.
That is irreplaceable.
It is the kind of data that rarely exists on the public internet in any structured form, and that AI systems moving into new domains increasingly need as they push into areas where enterprise telemetry, fitness records, and specialized imagery sit behind walls of ownership and consent.
The same logic applies to the richest social discussion platforms.
And then a dollar figure finally appeared.
The $60 million-a-year deal that just landed in the middle of a much bigger fight
In February 2024, Reddit signed a deal with Google that allows the search giant to use posts from the online discussion site for training its AI models, valued at roughly $60 million on an annualized basis, and now that agreement is ending soon, and the two companies are in talks about potentially renewing it.
Reddit stock dropped 9% on Wednesday after the Wall Street Journal reported the company had internally discussed ending Google’s ability to access its content for AI training purposes.
The reason Reddit is reconsidering is as telling as the deal itself.
With Google’s AI-powered query responses cutting into traffic to third-party sites, Reddit’s leadership is questioning whether the arrangement still serves the company’s interests.
A Pew Research study found users clicked traditional search links in about 15% of searches without an AI Overview, falling to 8% when one appeared, and just 1% for links inside the summary itself.
In other words: Reddit’s posts trained the tool that is now eating Reddit’s traffic.
Analytics data shows Reddit ranks among the most cited domains across major AI answer engines, including Google AI Overviews, Perplexity, and ChatGPT, so the arithmetic cuts both ways.
That leverage, analysts say, is worth far more than $60 million.
Wells Fargo forecasts combined licensing revenue could reach around $550 million upon renegotiation, versus a current run rate near $130 million.
For context on what this means for Google’s AI models, the data fueling them is no longer freely available, it now carries a price tag that is rising fast.
What comes next could reshape who owns the raw material of the AI era
The Reddit standoff is not an isolated contract dispute.
In its IPO prospectus filed in early 2024, Reddit disclosed data licensing arrangements with an aggregate contract value of $203 million, with deal terms ranging from two to three years.
Reddit also joined Yahoo, Medium, Quora, and other companies in backing Really Simple Licensing (RSL), which aims to standardize how AI developers license content and compensate publishers, with a collective licensing organization modeled after ASCAP for musicians.
The direction of travel is clear: the everyday content that millions of Americans create for free is being priced, negotiated over, and withheld, in a market that has gone from zero to billions of dollars in under three years.
It is a dynamic already visible in adjacent corners of the web. As GCN has reported, even a free tool used by 115 million Americans carries a hidden economic value that economists are only beginning to measure.
The Reddit deal puts a concrete dollar figure on something most users never imagined had a price.
And if Wells Fargo’s upside case holds, the next number will be even harder to ignore.
