AfterQuery — YC's Fastest Unicorn at $3.2B

AfterQuery, an AI training-data startup founded less than two years ago by a 22-year-old and a 23-year-old, reached a $3.2 billion valuation on September 1, 2026 — making it the fastest company in Y Combinator's nearly two-decade history to cross the unicorn threshold. The company's Series A closed just five months earlier at $300 million; that tenfold jump puts it in company with only a handful of enterprise startups that have ever moved this fast. The milestone was confirmed by YC partner Gustaf Alströmer, and multiple outlets reported it simultaneously.

The Claim

AfterQuery's founders, Carlos Georgescu and Spencer Mateega, entered YC's Winter 2025 cohort with a thesis that runs counter to how most people think about AI improvement. The dominant assumption in the industry is that smarter models come from more data and better answers — label outputs as correct or incorrect, reward the right ones, repeat. AfterQuery's pitch is that this is the wrong unit of analysis. What frontier labs actually need, the company argues, is data that encodes not whether an answer is right, but how a skilled practitioner would have arrived at it: the intermediate decisions, the reasoning chains, the moment-to-moment judgment calls that separate expert performance from adequate performance.

The company built its offering around a network of roughly 100,000 verified professionals — physicians, lawyers, software engineers, finance analysts — who generate training data and reinforcement learning environments by working through tasks in their actual domains. That data, AfterQuery claims, is structurally different from anything a model can learn by scraping the public web, because professional reasoning in clinical, legal, and financial contexts simply does not appear online in the form that makes it useful for training. A doctor's diagnosis is on the chart; the sequence of decisions that led there is not.

AI Weekly reports that AfterQuery's customer base includes OpenAI and Anthropic, alongside Nvidia, the legal AI firm Legora, and Korean AI lab Motif Technologies. The company has passed $100 million in annual recurring revenue and, according to multiple reports, is profitable — an unusual status for a startup at this stage and speed of growth. The competitive framing the company uses positions it against Scale AI, which built its business on labeling data for accuracy; AfterQuery is arguing for a different layer of the training stack, focused on process rather than correctness.

What We See

The "process training" framing is genuinely interesting and not obviously wrong. The limits of accuracy-only training have shown up repeatedly in the behavior of deployed models: they can state correct facts while failing at multi-step tasks that require sustaining a coherent line of reasoning across many decisions. If AfterQuery's expert network is actually generating data that captures that decision structure — not just the endpoint — that is a real product differentiation, not a marketing reframe.

Our read is that the moat claim deserves more scrutiny than the valuation alone suggests, but the underlying logic is sound. Professional behavioral data is difficult to collect at scale, expensive to generate, and impossible to fake convincingly with synthetic substitutes — at least at the current state of the art. The AI Weekly summary specifically characterizes AfterQuery's focus as coding and finance, which are two domains where that decision-sequence structure is both well-defined and in high demand from labs building code agents and financial reasoning tools. Cryptobriefing adds that the 100,000-professional network is verified, not crowdsourced — a distinction that matters if the value proposition is genuine expert judgment rather than volume.

The competitor named in multiple reports is Mercor, which operates in adjacent territory. Mercor's model is more explicitly focused on matching technical talent for AI evaluation work; AfterQuery's framing emphasizes process capture over talent supply. Whether those are genuinely distinct businesses at the level of what gets delivered to a lab's training pipeline is a question the public reporting does not resolve. The fact that customers include both OpenAI and Anthropic — the two labs most capable of building this kind of data pipeline themselves — is the strongest signal in AfterQuery's favor. Those are not customers who outsource data work because they can't do it; they outsource it when the external vendor's output is materially better than what they would build in-house.

Where It Falls Short

The $3.2 billion valuation is, at this point, mostly a claim about what sophisticated investors believe AfterQuery will be worth — not a verified measure of what it has already built. A tenfold jump in five months reflects investor conviction in the thesis, the team, and the revenue trajectory, but it also compresses a lot of risk into a number that will look either very right or very embarrassing depending on whether the data-quality advantages the company claims can be demonstrated rigorously at scale. The reporting across all three sources relies heavily on the company's own framing; there is no independent benchmark or third-party evaluation of whether models trained on AfterQuery data outperform those trained on alternatives in a controlled comparison.

There is also an open question about what happens to this model as frontier labs get better at generating synthetic expert data through their own models. The current moment favors human-in-the-loop expert annotation precisely because AI-generated reasoning data is not yet reliable enough to train on in high-stakes domains. If that changes — and labs are actively trying to make it change — the structural moat AfterQuery is describing narrows. None of the sources address this risk directly, which is notable for reporting on a company that just raised at a valuation implying a long and durable competitive position. The profitability figure is encouraging, but the pace of the valuation step-up suggests the real bet here is not on current economics but on whether the company can define and own a category before the window closes.

Sources