AI Training Data Startup AfterQuery Raises Valuation to $3.2 Billion, Founders Aged 22 and 23
AI training data startup AfterQuery has completed a new round of financing, with a post-money valuation reportedly reaching $3.2 billion, making it the fastest company in Y Combinator's history to become a unicorn from its founding.
The company announced in April that it had completed a $30 million Series A funding round, with a valuation of $300 million at that time; within five months, its valuation expanded more than tenfold, indicating that this round of capital has repriced its revenue growth and training data supply capabilities at a much higher rate than the previous round.
Y Combinator partner Gustaf Alströmer stated that AfterQuery set a YC record for the speed at which it surpassed a $1 billion valuation. The two founders are currently 22 and 23 years old, having only participated in the YC Winter 2025 batch 18 months ago.
AfterQuery disclosed in April that its annualized revenue run rate has reached $100 million; if this trend continues, the $3.2 billion valuation corresponds to about 32 times the annualized revenue, which falls within the high premium pricing range for high-growth AI infrastructure and data service companies.
The company has disclosed clients including NVIDIA, legal tech company Legora, and South Korean AI lab Motif Technologies. The client structure indicates that its revenue does not rely solely on a single model lab, but also covers chips, vertical software, and regional AI R&D institutions.
AfterQuery employs professionals such as doctors and lawyers to participate in model training, but its product goal is not merely to improve the accuracy of Q&A, but to encode the workflows, judgment paths, and reasoning patterns of professionals into the model and agent systems.
In market dynamics, buyers are large model labs, enterprise software companies, and computing power ecosystem participants that need to enhance AI agent task delivery capabilities; sellers are vendors capable of organizing scarce expert labor and converting implicit professional processes into trainable data. The financing event will continue to direct funds towards expert data recruitment, data production, and agent evaluation, putting pressure on similar service providers like Scale AI and Mercor in terms of talent and client competition, while chip companies, model companies, and enterprise software clients with large AI budgets will benefit from more executable agent capabilities.
Source: Public Information
ABAB AI Insight
AfterQuery is not entering the traditional crowdsourced labeling market, but rather the "high-value expert data" track validated by Scale AI and Mercor. Scale AI initially relied on structured data labeling services to penetrate the autonomous driving and machine learning client markets, later expanding its business to generative AI model evaluation, fine-tuning, and government contracts; Mercor focuses on recruiting knowledge workers to participate in AI training. AfterQuery further shifts the competitive focus to the execution process of professional tasks, moving from "Is the answer correct?" to "Does the model perform the work as professionals would?".
The key to the capital path is that the $100 million annualized revenue allows AfterQuery to no longer be just a training data project with unverified demand, but to become an AI infrastructure asset that can be valued by growth funds based on revenue multiples. The actual use of the new funds will typically concentrate on three fronts: attracting doctors, lawyers, and other scarce experts with higher compensation; establishing task decomposition, quality control, and reasoning trajectory collection systems; and expanding delivery teams targeting model companies and enterprise clients. Once clients integrate training data into core model iterations, the cost of replacing suppliers will increase as it becomes tied to evaluation standards, data formats, and expert networks.
The distinction from traditional data labeling companies lies in the training subjects shifting from static samples to executable workflows. Scale AI's value once came from industrializing data production; AfterQuery and Mercor are competing to productize the judgment, exception handling, and decision sequences within knowledge work. Similar shifts are also seen in the software industry: RPA first automates clearly defined repetitive operations, and subsequent AI agents attempt to handle complex tasks that require judgment and can tolerate errors across tools. AfterQuery is in the expansion phase of agents moving from demonstration capabilities to production deployment.
This is essentially a transfer of pricing power: as model parameters and generic web text become increasingly commoditized, the scarce professional data that can enhance real task completion rates becomes the new bottleneck. Model companies can buy GPUs, rent cloud computing power, and access large-scale public corpora; however, the diagnostic processes of doctors, the review paths of lawyers, and the experience of handling exceptions in corporate roles cannot be easily obtained through simple web scraping. The party that controls the supply of experts, quality screening, and task standards will gain stronger bargaining power over training budgets.
ABAB News · Cognitive Laws
Computing power determines the ceiling, process data determines the landing
Answers can be replicated, but judgment paths are harder to purchase
In the end, model competition is about competing for the interface of human experience