An AI Training Data Startup Just Became Y Combinator’s Fastest-Ever Unicorn

AfterQuery, an 18-month-old AI startup, has reached a $3.2 billion valuation in a new funding round, Forbes reported Monday. That figure is more than ten times the $300 million valuation it carried five months earlier, when it closed a $30 million Series A—an early funding round where a startup sells a slice of itself for growth capital.

The jump makes AfterQuery the fastest startup in Y Combinator’s history to go from inception to unicorn status, according to YC partner Gustaf Alströmer. AfterQuery declined to comment on the report. One of Forbes‘ sources said the company is already profitable and has lined up a lead investor for the round.

Spencer Mateega and Carlos Georgescu founded the company in February 2025, 18 months after joining Y Combinator’s Winter 2025 batch as two high school friends with no product and no fixed idea. Mateega posted on X in July that annual recurring revenue—the yearly value of a company’s ongoing subscriptions—had grown to the “hundreds of millions,” up from $100 million in April.

The founders originally wanted to build AI agents for finance. Testing showed leading models kept failing at nuanced, professional-grade calls—not from lack of raw capability, but because no one had ever taught them how an expert actually reasons through a hard decision.

So they pivoted. AfterQuery now pays specialists to produce reasoning data: written, step-by-step records of how a professional works through a problem, used to teach AI models judgment instead of just facts. Nvidia has used that data to train its open-source Nemotron models, and AfterQuery also counts former OpenAI CTO Mira Murati’s Thinking Machines Lab and legal AI firm Legora as clients, per Forbes.

The demand behind that pivot is industry-wide. Frontier labs—the companies building the most advanced AI systems—have already picked through most of the usable text on the open web, and synthetic data, or AI-generated training material, only goes so far. What’s left in short supply is judgment that only comes from real, credentialed experts.

Other companies are chasing the same shortage from different directions. South Korean fintech Toss recently opened its 30 million users to the AI data economy through a partnership with data infrastructure firm Poseidon, paying ordinary users to record real-world data that models can’t find online.

AfterQuery isn’t the only one cashing in on the shortage. Scale AI’s Alexandr Wang became the industry’s first data-labeling billionaire in 2021 at age 24, before Meta paid $14.3 billion for a 49% stake and put Wang atop its own AI lab. Rival Mercor’s founders passed Wang’s early record last October, becoming billionaires at 22.

Mateega has said AfterQuery’s edge over Mercor, which leans on an AI interviewer to staff a large contractor pool, is custom software that screens submissions for a “Goldilocks” difficulty—hard enough to challenge a frontier model, not so hard it can’t learn from the answer. AfterQuery also trains its own models on the data before selling it, to show labs the material actually moves the needle rather than asking them to take its word for it.

Mercor, for its part, is currently in talks with Nvidia for a funding round that would value it at $20 billion, doubling its $10 billion price tag from last October. AfterQuery’s round has yet to close, and the company has not named its lead investor.