There's a weird thing happening with Reddit stock, and if you blink you'll miss the real story. The market looks at RDDT and sees a social media company — ad revenue, user growth, the standard playbook. That's not wrong, exactly. But it's dangerously incomplete. What the market hasn't fully grasped is that Reddit has quietly transformed into something far more valuable: the internet's essential training ground for artificial intelligence. And that changes everything about how you should think about this company.

The numbers on the surface already tell a strong story. Reddit pulled in roughly $2.47 billion in trailing twelve-month revenue, up 69% year over year. Net income hit $529.7 million in fiscal 2025 — the first GAAP-profitable year in company history. Gross margin sits at a ridiculous 91.4%. Operating margin exploded from basically nothing — 1% in Q1 2025 — to 27.6% on a TTM basis. Free cash flow is $564 million. The company has $2.77 billion in cash against a microscopic $21.3 million in debt. By any traditional measure, this is a business that has figured out how to make money.

But here's where it gets interesting. Reddit's ad business — which still accounts for 94% of revenue — is firing on all cylinders. Q1 2026 ad revenue hit $625 million, up 74% year over year. US ARPU climbed to $9.20, a 61% jump. Daily active unique users hit 126.8 million in Q1, growing 17% year over year. The engagement metrics are absurd: revenue per engagement hour doubled to $0.34. Their Conversation Placement ads deliver a 0.41% click-through rate, more than double what the home feed does. Reddit Max, their new AI-powered ad product, hasn't even fully ramped yet.

That's the story Wall Street sees. It's a good story. It's just not the biggest one.

The real story is sitting in Reddit's data licensing business, and it's about to get a lot bigger. Right now, Reddit has deals with Google — roughly $60 million a year — and OpenAI at about $70 million. Combined, that's around $130 million annually, or maybe 6% to 10% of total revenue. Not nothing, but not game-changing. Except that Wells Fargo analysts have penciled in a scenario where renegotiations push that number to $550 million a year. That's more than four times the current run rate. And there's good reason to think they could get there.

Consider this: Reddit is the single most-cited domain in AI-generated answers across Google AI Overviews and Perplexity. It's the second most-cited source in ChatGPT. A full 5.5% of all AI Overview replies pull from Reddit somehow. That's not an accident. It's not nostalgia. It's structural. Reddit has more than a billion posts — an irreplaceable corpus of real human conversation, argument, advice, and knowledge that no synthetic data set can replicate. When an AI needs to know how to fix a dishwasher, what it actually feels like to quit a job, or which hiking trail is actually worth the drive, it learns from Reddit. There is no substitute.

The company knows this. Instead of sitting back and collecting checks, Reddit is actively defending its data moat. They're suing Anthropic for scraping without permission. They're negotiating dynamic pricing models where the fee scales with the value their data generates in AI outputs, not some flat annual number. They're treating their data like the strategic asset it is, not a side hustle.

The bull case here is elegant. If data licensing hits $500 million-plus annually — and the trajectory of AI adoption suggests that's conservative — Reddit's earnings profile shifts dramatically. At the current $38 billion market cap, that would put the stock trading at roughly 30 times net income instead of the 57 times trailing earnings it commands today. The market is pricing RDDT like a cyclical ad business. It's actually building a toll road between the human internet and the machine-learning economy.

Of course, there are risks. The ad business is still the engine, and any macro slowdown hits that first. The data licensing revenue is still small enough that a renegotiation flop would sting. And the AI landscape shifts fast — tomorrow's models might not need today's training data the same way. But structural scarcity is a hell of a negotiating position. There are roughly 20 companies building frontier AI models, and exactly one of them owns a billion-plus archive of real human conversation.

Reddit went from being the front page of the internet to being the internet's data layer for AI. The market hasn't caught up to that story yet. But it will.

Disclosure: The Signal holds no position in RDDT. Positions may change. This is not financial advice.