A startup that barely existed 18 months ago is reportedly worth $3.2 billion.

And it isn’t building another chatbot.

AfterQuery is selling something arguably more valuable to the companies building those chatbots: examples of how skilled humans actually think, make decisions, use software and finish complicated work.

On September 1, Forbes reported that AfterQuery had reached a $3.2 billion valuation, more than 10 times the $300 million valuation it received only five months earlier. A Y Combinator partner described it as the fastest startup in the accelerator’s history to reach unicorn status. The company has not publicly confirmed the new round, however, and its size and lead investor remain undisclosed.

The money is eye-catching. The more interesting question is whether AfterQuery has found one of the genuine bottlenecks holding back the next generation of AI.

So what exactly is AfterQuery—and does a company this young really deserve a multibillion-dollar valuation?

What’s Going On With AfterQuery?

AfterQuery has gone from obscure Y Combinator startup to one of the hottest companies in AI infrastructure remarkably quickly.

Its April 2026 Series A already looked unusual. AfterQuery’s $30 million Series A announcement said the company had raised $30 million at a $300 million valuation, led by Altos Ventures with participation from The Raine Group, Y Combinator, BoxGroup and Latitude Capital. At the same time, the company said it had surpassed a $100 million annualized revenue run rate only 14 months after inception.

By July, CEO Spencer Mateega was describing recurring revenue as being in the “hundreds of millions,” according to Forbes. Forbes’ September report also cited a source saying the company was profitable. Those are extraordinary figures for such a young business—but they remain private-company figures rather than audited public financials.

Then came the reported $3.2 billion valuation.

What’s driving that acceleration is a broader shift in AI development.

Frontier models have already consumed huge quantities of text, code and other information from the open internet. Improving them on difficult professional work increasingly requires something else: carefully designed tasks, realistic environments, expert judgments, detailed reasoning traces and feedback that shows models not merely what the answer is, but how competent professionals get there.

That’s the market AfterQuery is chasing.

What Is AfterQuery?

AfterQuery is an AI training-data and post-training infrastructure company based in San Francisco.

Its simplest pitch is that it helps AI models learn how experts work.

Instead of paying thousands of generic annotators to label whether an answer is good or bad, AfterQuery recruits professionals in fields including software engineering, finance, medicine and law and turns their expertise into training and evaluation material.

The company’s current product portfolio includes:

• supervised fine-tuning datasets and reasoning traces;
• reinforcement-learning datasets, rubrics and verifiers;
• tool-calling environments built around APIs and MCP;
• computer-use trajectories;
• custom model evaluations and benchmarks;
• professional-domain datasets;
• post-training and enterprise AI implementation work.

That makes AfterQuery broader than a conventional “data labeling” business.

It is increasingly trying to occupy the entire loop between identifying where an AI system fails, constructing realistic tasks around those failures, generating expert training data, validating it, post-training the model and measuring whether performance actually improves.

There isn’t really a consumer AfterQuery product to download and play with. This is predominantly business-to-business infrastructure sold to AI labs and enterprise teams.

Who Founded AfterQuery?

AfterQuery was founded in early 2025 and joined Y Combinator’s Winter 2025 batch.

The two founders most prominently associated with the company are Spencer Mateega, CEO, and Carlos Georgescu, CTO. Forbes’ original profile of AfterQuery reports that the pair met while participating in a Google computer-science program in high school and later overlapped as interns at Meta. Mateega studied finance and statistics at Wharton alongside computer science at the University of Pennsylvania, while Georgescu studied computer science at the University of British Columbia and had engineering experience at companies including Meta, Google and Citadel Securities.

There is a small wrinkle in the founder record worth noting. AfterQuery’s own About page also lists Danny Tang as a cofounder, and AfterQuery’s Y Combinator profile discusses Tang as part of the founding team, even though most recent press coverage frames Mateega and Georgescu as the company’s founders.

The original idea was reportedly quite different.

Mateega and Georgescu initially intended to build AI agents for finance. They discovered that existing models struggled with the nuanced judgment required in professional workflows. Instead of building another application on top of those models, they pivoted toward producing the data needed to make the underlying models better at the work.

That pivot may have been the most important decision the company made.

Why Is AfterQuery Suddenly Getting So Much Attention?

Three things stand out.

1. The growth numbers are extraordinary

AfterQuery went from a $300 million valuation in April to a reported $3.2 billion valuation five months later.

It also went from announcing a $100 million-plus annual revenue run rate in April to Mateega saying recurring revenue was in the “hundreds of millions” by July.

Those numbers alone would get investors’ attention.

They should also be treated carefully. AfterQuery is private, and the company does not publish audited accounts.

2. Its product is showing up in serious AI work

This matters more.

NVIDIA Nemotron 3 Ultra technical report explicitly describes constructing a training distribution using AfterQuery tasks involving file-grounded reasoning, professional deliverables, multi-step analysis and judged final outputs. That is considerably stronger evidence than a customer logo on a startup website.

AfterQuery also announced in August that AfterQuery’s work on Motif 3 served as Motif Technologies’ sole data partnership for the Korean lab’s Motif 3 release.

Then, on August 30, it disclosed a collaboration with legal AI company Legora to build AfterQuery and Legora’s BAR legal benchmark. BAR contains 5,161 cases across 28 legal practice areas and 11,075 source documents. AfterQuery says changes informed by the benchmark improved output quality across the same production models by roughly 5% after a month of use.

That is what makes the AfterQuery story more interesting than a fundraising story.

Customers appear to be using the work in production-oriented model development.

3. AI’s bottleneck is moving

The easiest training data has already been collected.

Now AI companies are competing over difficult data: expert reasoning, long-horizon tasks, software interaction, realistic agent environments and verifiable professional work.

AfterQuery happens to have arrived precisely as that transition is accelerating.

How Does AfterQuery Work?

The basic AfterQuery loop looks something like this:

AI lab identifies a weakness → AfterQuery designs realistic tasks → domain experts complete the work → software and research systems validate the data → model is fine-tuned or trained with reinforcement learning → performance is evaluated → the process repeats.

The important part is validation.

AfterQuery argues that relying primarily on human reviewers creates an obvious problem: difficult expert tasks are themselves difficult to grade. A plausible-looking response may still contain subtle mistakes, and a task that is too easy teaches the model little.

The company’s stated approach is therefore “software-first.” It builds custom systems for generating, checking and scoring expert data and tries to create tasks in a useful difficulty range—not so easy that a model already solves them, and not so impossible that the feedback becomes noise.

AfterQuery is also increasingly focused on environments, not just static datasets.

An AI coding agent, for example, may need a repository, terminal and tools. A finance agent may need spreadsheets, files and APIs. A legal agent may need a document corpus and a realistic matter to investigate.

AfterQuery can construct those environments and pair them with tasks, expected behavior, rubrics or verifiers so models can learn through reinforcement learning.

Crucially, AfterQuery is not primarily building its own frontier foundation model. Its role is closer to the specialized training layer sitting between human expertise and the model companies.

What Can You Actually Do With It?

For an AI lab, a project might look like:

Train a coding agent: Give expert software engineers realistic repositories and bug reports, record successful tool-use trajectories, validate the changes and use those examples to improve an agent. For more context, see our guide to AI models for coding.

Improve a legal agent: Build a simulated legal matter containing thousands of pages of source material, define what a competent lawyer should investigate and produce, then measure whether the agent completes the workflow correctly.

Teach models office work: Have professionals complete spreadsheet, document-analysis and file-grounded tasks so a model learns to produce the kind of deliverables people actually create at work.

Build an RL environment: Create APIs, tools, state changes and scoring rules that let an AI agent repeatedly practice a workflow instead of learning from one static answer.

Evaluate an enterprise model: Design a benchmark around a company’s actual workflows, find the system’s weak spots, build targeted training material and test whether post-training improved them.

The last example is particularly important because AfterQuery is moving beyond selling datasets into evaluation, post-training and deployment services through its enterprise AI solutions.

AfterQuery Pricing

AfterQuery does not publish conventional self-serve SaaS pricing.

Its customer-facing pages direct buyers to request datasets, samples, custom pilots or enterprise engagements, while its AfterQuery terms of service state that fees are established through separate agreements or order forms.

Offering	Current Public Pricing
Off-the-shelf datasets	Contact sales
Custom datasets	Custom quote
Evaluations/benchmarks	Custom engagement
RL environments	Custom quote
Post-training	Custom engagement
Enterprise AI implementation	Contact sales
Free customer tier	None publicly advertised
Public API pricing	Not published
Public free trial	None found

Don’t confuse that with AfterQuery Experts, the company’s contributor network. AfterQuery advertises paid project work for specialists, with rates varying significantly by profession and project. Those are payments to expert contributors, not product prices for AfterQuery customers.

The Numbers Behind the Hype

Metric	Reported Figure
Founded	February 2025
Headquarters	San Francisco
YC batch	Winter 2025
Publicly announced funding	At least $30M
April 2026 Series A	$30M
April valuation	$300M
Latest reported valuation	$3.2B
April revenue run rate	$100M+ — company-reported
July recurring revenue	“Hundreds of millions” — founder statement reported by Forbes
Profitability	Reported by Forbes source; not company-confirmed
Current expert network	Company site currently claims 300,000+ verified practitioners, although another company page still cites nearly 100,000

Sources: AfterQuery, Forbes, TechCrunch and Y Combinator.

That last number deserves an asterisk.

AfterQuery’s current Products page says its network spans 300,000+ verified practitioners, while its About page still refers to nearly 100,000. The latter appears to reflect an older April figure, but the discrepancy means the 300,000 figure should be described explicitly as a current company claim rather than a clean independently verified user count.

What Makes AfterQuery Different?

The obvious answer—”better data”—isn’t enough.

Every company in this market says it has better data.

AfterQuery’s more credible differentiation is its attempt to combine four things:

High-end experts. It recruits professionals capable of producing difficult domain-specific work rather than limiting itself to generic annotation.

Software-based validation. The company emphasizes custom systems for checking and shaping the data rather than making human review its only quality-control layer.

Research and post-training. AfterQuery doesn’t simply hand a customer a dataset. It runs internal experiments designed to demonstrate that the data changes model performance.

Realistic agent environments. It is positioning for the shift from chatbots toward agents that interact with files, software and tools.

The NVIDIA work is important because it provides third-party evidence that this approach is useful to a serious model developer.

But none of those advantages is guaranteed to become a permanent moat.

Competitors are moving in exactly the same direction.

Who Is AfterQuery Competing Against?

Company	Positioning	Main Difference vs. AfterQuery
Mercor	Expert talent, model-training projects, datasets and RL environments	Strong marketplace/recruiting infrastructure and rapid scaling of professional contributors
Scale AI	Broad AI data infrastructure, evaluations, RLHF and RL environments	Far larger incumbent with established enterprise and government relationships
Turing	Large global technical/expert network plus frontier-model training data	Enormous contributor network and long-standing engineering talent business
Handshake AI	Professional/academic expert network for AI research and training	Can leverage Handshake’s existing education and professional distribution
AfterQuery	Research-driven expert data, evaluations, RL environments and post-training	Emphasizes software validation and closing the loop from data creation to measured model improvement

Mercor AI research and training data shows how Mercor is positioning itself around expert-built datasets and AI research. Scale AI reinforcement-learning environments demonstrate Scale’s focus on RL environments, while Turing Frontier AIhighlights Turing’s frontier-model data and expert network.

So AfterQuery isn’t winning because nobody else saw the opportunity.

It is trying to win on execution.

What People Like About AfterQuery

Enterprise customer reviews are still sparse, which makes it difficult to claim broad customer consensus.

The strongest positive signal is therefore not a five-star review. It’s adoption.

NVIDIA documented AfterQuery tasks inside its training work. Legora collaborated with the company’s research team on a substantial legal benchmark. Motif named AfterQuery’s role in its latest model work. Those examples suggest AfterQuery can produce data that sophisticated AI teams consider useful.

Among experts doing paid project work through AfterQuery, positive comments tend to focus on flexibility, relatively high advertised rates and the ability to contribute to advanced AI projects.

There are also reports from contributors saying they were paid successfully.

But contributor sentiment is decidedly mixed.

What People Don’t Like

The most visible complaints involve AfterQuery Experts, the supply side of the business.

As of early September, AfterQuery reviews on Trustpilot show a mixed rating and reviews complaining about delayed task approval, payments taking longer than expected, projects suddenly becoming unavailable and difficulty getting useful support. Other reviewers report positive experiences and successful payments. Trustpilot reviews are user-submitted and cannot establish that any individual allegation is true.

Reddit discussions show a similar pattern: some contributors report receiving payment, while others describe Stripe holds, extended reviews or confusion around projects and account status. Those comments are anecdotal, but the recurrence of operational complaints is worth watching because AfterQuery’s product depends on maintaining a large pool of high-quality experts.

A second weakness is customer concentration risk.

Frontier AI labs spend enormous amounts on training data, but there are relatively few of them. A vendor can grow extraordinarily quickly when it wins a major lab—and see that revenue move just as quickly if a customer changes suppliers or brings more work in-house. Forbes has previously highlighted how competitive and volatile the AI-data vendor market can be.

And then there is China.

The China Question

In August, Forbes’ investigation into U.S. AI-data vendors serving Chinese labs reported that AfterQuery and other American AI-data businesses were selling training services to Chinese AI companies.

Forbes reported, citing people familiar with the business, that AfterQuery generated at least $50 million in recurring revenue from Chinese AI labs, while communications reviewed by the publication indicated relationships involving Ant Group and Alibaba. AfterQuery declined to disclose customer information.

There is nothing inherently scandalous about selling lawful services internationally.

The issue is geopolitical.

Washington has spent years trying to restrict China’s access to advanced AI hardware. If American companies are simultaneously helping Chinese laboratories improve their models using high-quality American training infrastructure, lawmakers may eventually decide that data and post-training services deserve similar scrutiny.

That makes AfterQuery’s China exposure not merely a reputational question, but a potential regulatory and revenue risk.

Privacy and Security

AfterQuery works with exactly the kind of material that can become sensitive: enterprise files, professional workflows, customer data and custom agent environments.

Its AfterQuery privacy policy, updated in August 2026, says the company collects information such as contact and business details and uses technical and organizational measures to protect personal data. It also allows service providers to process information for operational purposes.

But the public documentation is not especially detailed.

The policy explicitly says personal information processed on behalf of customers falls outside the scope of the general privacy policy and is governed by the customer’s own responsibilities and arrangements. On the public pages reviewed for this article, AfterQuery does not spell out a general data-retention timetable, and I did not find a prominent public explanation of whether customer data may be used for model training.

I also would not infer enterprise-grade certification simply from the company’s customer list. A serious buyer should ask AfterQuery directly about:

• SOC 2 and ISO 27001 status;
• encryption in transit and at rest;
• data residency;
• retention and deletion;
• subprocessors;
• employee and contractor access;
• incident-response procedures;
• whether customer data can train models;
• contractual restrictions on reuse.

That doesn’t mean AfterQuery is insecure. It means its publicly available security detail is not sufficient to declare it secure on behalf of a customer.

Is AfterQuery Actually Legit?

Yes—in the basic sense of the word.

AfterQuery is a real company with identifiable founders, Y Combinator backing, a $30 million Series A led by Altos Ventures and documented technical work involving major AI companies. Most convincingly, NVIDIA’s own technical material references AfterQuery data rather than leaving the relationship as an unverifiable logo on a vendor page.

That’s strong evidence that this is a substantive AI infrastructure business.

But “legit” does not mean:

• every contributor will have a good experience;
• the reported $3.2 billion valuation is sensible;
• the company is risk-free;
• its revenue is durable;
• its privacy practices meet every enterprise requirement.

Those are separate questions.

AfterQuery Pros and Cons

Pros	Cons
Third-party technical validation from NVIDIA	$3.2B valuation remains reported rather than company-confirmed
Extremely strong reported revenue growth	Private financial figures are not independently audited
Positioned in a genuine frontier-AI bottleneck	Fierce competition from Scale, Mercor, Turing and others
Covers data, evals, RL environments and post-training	Contributor complaints around reviews, support and payments
Named work with NVIDIA, Legora and Motif	Limited public pricing transparency
Software-first approach to data validation	Public security/compliance information could be more detailed
Strong timing as AI shifts toward agents	China exposure could create regulatory risk

So… Does AfterQuery Deserve the Hype?

Mostly.

The reason isn’t the $3.2 billion headline.

If AfterQuery were merely another company recruiting cheap annotators and reselling their work at enormous margins, the valuation would be much harder to take seriously.

Instead, the company appears to be positioned around a real change in how advanced models are being developed.

The frontier is moving from “How much internet data can we feed the model?” toward “How do we teach a model to perform difficult, verifiable work inside realistic environments?”

That transition favors companies capable of combining expert humans, research, software, evaluations and reinforcement-learning environments.

AfterQuery has also produced something young startups rarely have: external technical evidence. NVIDIA actually describes using its tasks. Legora’s benchmark collaboration is specific. Motif’s relationship is specific.

That materially strengthens the case.

But a $3.2 billion valuation prices in a lot of future success.

The expert-data market is crowded. The biggest customers have leverage. AI labs can switch vendors, acquire suppliers, develop internal data operations or generate increasingly useful synthetic data. Scale, Mercor, Turing and Handshake are all pursuing overlapping opportunities.

AfterQuery also has operational work to do. Its contributor network is a core input into the product, so persistent complaints about task review and payment reliability cannot simply be dismissed as a side issue. And its exposure to Chinese AI labs could become much more consequential if U.S. policy expands from chip restrictions into training-data services.

My read: the underlying business deserves the attention more than the valuation does.

AfterQuery appears to have found a genuine AI infrastructure bottleneck and moved into it at exactly the right time. That’s meaningful.

Whether it becomes a durable platform—or merely one of several extremely valuable intermediaries in a temporary spending boom—is the harder question.

Who Should Try AfterQuery?

AfterQuery is most relevant for:

1. Frontier AI labs needing difficult SFT, RL or evaluation data.
2. Agent developers building systems that use software, APIs, files and tools.
3. Enterprise AI teams that need benchmarks based on their actual workflows.
4. Legal, finance, healthcare and engineering AI companies requiring domain experts rather than generic annotators.
5. Research teams trying to improve measurable model performance on professional tasks.

Small companies looking for a cheap self-serve API or ordinary SaaS tool should probably look elsewhere.

And experts considering contributor work should read the current project, review and payment terms carefully rather than assuming advertised hourly rates translate into guaranteed work.

FAQ

What is AfterQuery?

AfterQuery is an AI data and post-training company that creates expert-generated datasets, reinforcement-learning environments, evaluations and related infrastructure for AI labs and enterprises.

Who founded AfterQuery?

Recent media coverage focuses on CEO Spencer Mateega and CTO Carlos Georgescu. AfterQuery’s own website and YC material also identify Danny Tang as a cofounder.

How much funding has AfterQuery raised?

AfterQuery publicly announced a $30 million Series A in April 2026. A newer financing has been reported, but its size has not been publicly disclosed. It therefore has at least $30 million in confirmed announced funding.

What is AfterQuery worth?

Forbes reported on September 1, 2026 that a new financing valued AfterQuery at $3.2 billion. AfterQuery declined to comment, so the figure should currently be described as a reported valuation rather than a company-confirmed one.

Is AfterQuery free?

There is no publicly advertised free customer plan. AfterQuery sells datasets and custom enterprise work through a sales process and does not publish standard SaaS pricing.

Is AfterQuery legit?

Yes. It is a YC-backed company with known investors and documented work in NVIDIA’s model-training research. That does not mean every claim, valuation or contributor experience should be accepted without scrutiny.

What are the best AfterQuery alternatives?

The closest competitors include Mercor, Scale AI, Turing and Handshake AI, depending on whether the customer needs expert data, RL environments, evaluations or a large specialist workforce.

Final Take

AfterQuery is interesting because it is getting rich from a problem the AI industry spent years pretending would disappear: humans still matter enormously to making better AI.

Models may generate synthetic training examples. Agents may eventually evaluate one another. But when the objective is to teach a system what excellent legal analysis, engineering work, financial judgment or complex software use actually looks like, high-quality human expertise remains unusually valuable.

AfterQuery has turned that requirement into one of the fastest-growing businesses in Y Combinator history.

Now the difficult part begins.

Watch whether the reported $3.2 billion financing gets formally confirmed, whether revenue continues expanding after the current post-training boom, whether more independent model reports document measurable AfterQuery-driven gains, how the company handles its contributor network, and whether Washington decides that supplying advanced training data to Chinese AI labs requires new restrictions.

If AfterQuery keeps producing NVIDIA-style proof while doing all of that, the company may matter far more than its relatively low public profile suggests.

Would you bet that expert training data becomes one of AI’s most valuable infrastructure markets—or do you think models will eventually learn to generate enough of it themselves?