The new artificial intelligence gold rush is not just about GPUs, data centers, and models. It is increasingly about the quality of human expertise being turned into data. AfterQuery, a startup founded by Spencer Mateega and Carlos Georgescu that went through Y Combinator's Winter 2025 batch, has reportedly reached a valuation of $3.2 billion in a new funding round reported by Forbes. TechCrunch picked up the news, highlighting the exceptional pace of growth: just five months earlier, the company had announced a $30 million Series A at a $300 million valuation.
AfterQuery has not publicly disclosed all the terms of the new round and had not immediately responded to requests for comment cited by the press. It is therefore important to refer to a reported valuation, not a figure directly announced by the company. If confirmed, it would make the company one of the fastest startups to reach unicorn status in Y Combinator history.
Teaching the model to respond is no longer enough
AfterQuery belongs to a new generation of companies recruiting professionals—doctors, lawyers, engineers, and other specialists—to produce highly complex work samples. The difference compared to early data labeling services is substantial. It is not about indicating whether a photo contains a cat or a dog, but about capturing sequences of decisions, reasoning, and actions that describe how an expert completes a task.
The company summarizes this activity as encoding the patterns and decisions of top professionals. The goal is to train not only models that know information, but agents capable of tackling long processes and using tools in a manner similar to a competent worker.
Easy benchmarks are losing value
The best models have now achieved high scores across many standardized tests. Once a benchmark becomes saturated, it ceases to effectively distinguish between systems. Labs therefore need harder, more realistic, and verifiable problems, often built with the help of specialists.
This creates demand for expensive data. An hour of a doctor's or senior engineer's time is worth far more than an hour of generic annotation, but it can produce examples that distinguish a model capable of formulating a plausible answer from one capable of actually completing a job.
The reported growth is impressive, but needs to be put into context
In April, AfterQuery reported an annualized revenue run rate of $100 million, naming clients such as Nvidia, Legora, and Korean lab Motif Technologies. A run rate does not equate to actual revenue already collected and can shift rapidly, especially in a nascent market. However, it remains an indicator of the amount of money labs are pouring into improving post-training, evaluation, and agentic behavior.
The $3.2 billion valuation assumes that this demand will remain strong and that AfterQuery can defend its margins even as clients develop in-house pipelines or models become more adept at generating synthetic data.
Human labor is not disappearing: it is shifting its position in the chain
A common narrative holds that AI will reduce the need for skilled professionals. AfterQuery's business reveals a paradox: building systems capable of automating skilled work requires vast amounts of skilled labor in the first place. Doctors, legal experts, and engineers are becoming teachers to the models.
The economic question is how long this phase will last. Synthetic data can expand datasets and simulate experiences, but verifying whether an output is correct often requires an external point of reference. In highly complex domains, human experts remain the most reliable source of ground truth.
Scale AI proved the value of data infrastructure
AfterQuery is following the trail blazed by companies like Scale AI and Mercor. The difference is that the market is moving up the value chain. Basic labeling is becoming easier to automate, while value is concentrating in tasks that require specialized judgment, access to scarce professionals, and quality control systems.
This is pushing startups to build not just software, but genuine marketplaces for cognitive labor. Recruiting, skill verification, and quality management are becoming core parts of the product.
Labs can be both customers and competitors
Major AI companies have enough capital to build internal data operations teams. At the same time, they often prefer to procure external capacity to scale quickly or tap into niche professional talent. AfterQuery must therefore deliver enough value to remain strategic even as its clients build out their own internal capabilities.
The advantage could lie in the network: more qualified professionals, more proven processes, and a greater capacity to produce datasets quickly. But the barriers are not absolute, and the market could consolidate.
Data quality becomes a governance issue
Teaching a model “how a professional works” is not neutral. Different professionals may make different decisions, and every dataset incorporates criteria, practices, and assumptions. Companies building training data must therefore document who produces the examples, how quality is evaluated, and which divergences are deemed acceptable.
In regulated sectors, this aspect will be crucial. A model trained on medical or legal workflows cannot be treated as though it had learned a single universal truth.
A massive valuation for an infrastructure still in its infancy
AfterQuery shows just how much the AI market is willing to pay for bottlenecks that seemed secondary just a few years ago. As models grow, compute is not the only scarce resource; tasks difficult enough to teach something new and the experts capable of evaluating them are also becoming scarce.
The $3.2 billion valuation, if the round is confirmed on the reported terms, is a bet that this scarcity will persist. It is possible that models will learn to autonomously generate an increasing portion of their own training. It is equally possible that each leap in capabilities will create a need for even more sophisticated examples. AfterQuery has built its business on the second hypothesis.
