> ## Content Index > Fetch the complete content index at: https://www.signalful.com/llms.txt > Use this file to discover other available public pages before exploring further. # AI data companies should run like hedge funds - URL: https://www.signalful.com/ai-data-companies-hedge-funds/ - Published: 2026-11-09T14:00:00.000Z - Updated: 2026-11-10T15:13:39.000Z - Description: Use an S-Corp, skip VC, pay a share of gross profit instead of equity, set aside taxes, reinvest 30% and distribute the rest. A recruiting firm already runs this way. - Author: Gokul Rajaram - Tags: Builders, Business Model, Company Building, #analysis, Analysis By Gokul Rajaram I've met with 50+ AI data companies over the past few weeks. Without doubt, it's one of the most prolific and interesting segments within AI. I'll first explain what these companies do, why they matter in the world of AI, and then articulate my thesis around how they should be organized and run. ## What exactly do AI data companies do? Frontier AI labs such as OpenAI and Anthropic need more than Internet text to train their models. In order to go beyond simple content generation and tackle problems in complex real-world domains, these AI models need increasingly specialized, high-quality data from experts who can solve hard problems, judge whether an answer is correct and show the models what good work looks like in medicine, law, coding and finance. AI data companies recruit domain experts and turn their knowledge into training examples, evaluations and reinforcement learning (RL) environments. They handle task design, expert matching, quality control and delivery to the lab. Simply put, **AI data companies help AI labs make their models better by supplying the examples, feedback and practice tasks that teach them to do useful work.** The growth of the leading companies in AI data has been extraordinary, to say the least. - **Mercor** supplies domain experts to frontier labs and also builds RL environments in several domains (eg coding, legal, etc). It crossed $1B in annualized gross revenue in early 2026, doubling from September 2025. - **micro1** builds expert datasets and training environments. It crossed $1B in annualized revenue in October 2026, up from $500M in summer 2026. - **Surge AI** provides high-quality labeling and human feedback to labs including Anthropic and OpenAI. Founded in 2020, it reportedly exceeded $1B in annual revenue in 2024, importantly while staying bootstrapped and profitable. - There are at least half a dozen other AI data companies that have gone from 0 to $100m+ annualized revenue in less than a year. The Anthropic chart below shows the spending behind this growth. Anthropic’s annualized data vendor spend rose from roughly $40M in December 2023 to $530M in December 2025, $900M in June 2026 and $2.4B in September. That is a 60-fold increase in under three years. ![](https://storage.ghost.io/c/80/32/80323665-8fd8-4f6a-8b1d-18cf5988ae6d/content/images/2026/11/image.png) Anthropic, OpenAI, Google DeepMind, Meta Superintelligence Labs, and a handful of other frontier labs are spending billions of dollars to buy the expert knowledge that improves their next model. Their expanding budgets can turn a small supplier with trusted experts and reliable delivery into a very large business quickly. The same mechanism also concentrates revenue: a data company’s growth depends heavily on the priorities and spending of a handful of customers. Inspired in particular by Surge, and how they've surged (no pun intended) to $1B in annualized revenue with zero external capital raised, I've been thinking about how AI data companies should optimally be set up: the legal entity, funding, compensation structure and organization design. To summarize, I believe most AI data companies should operate like a multi-manager hedge fund. Rough thoughts below. **This is not legal or tax advice.** Talk to a lawyer and an accountant before you file anything. ## Why AI data companies are like hedge funds Three things make an AI data business look like a hedge fund. **Their edge decays with time.** A fund's alpha shrinks as other people find and enter the same trade. A data company's edge decays too. Labs’ needs move from generalist labeling to expert reasoning, coding tasks and RL environments. Suppliers must keep finding the next valuable capability. **Profit is clearly attributable to people and teams.** At a hedge fund, you know P&L for each Portfolio Manager. At a data company, you can track revenue and delivery costs by lab, project and team. That makes team-level profit sharing practical. **It needs relatively little capital to get to positive cash flow.** The main inputs are people, tooling and quality control. Once customer revenue covers those costs, growth depends heavily on recruiting experts and earning lab trust, which take relatively little capital (since experts are paid based on a share of revenue generated). ## How I'd structure an AI data company The best comparison for AI data companies is to the hedge fund management company, whose operating capital is separate from the money investors entrust it to invest. Here’s how I’d apply that model. ### 1\. Use an S-Corp or LLC If the business is built to generate and distribute cash, choose a structure that makes that efficient. A C-Corp pays corporate income tax, and shareholders pay tax again when profits are distributed as dividends. An S-Corp or an LLC taxed as a partnership generally passes income through to its owners, avoiding that second layer of federal income tax. The tradeoff is flexibility. S-Corps restrict who can own shares and allow only one class of stock. Owners who work in the business must receive reasonable compensation. An LLC offers different flexibility and administrative requirements. Work with your lawyer and accountant to choose between them. ### 2\. Skip venture money, at least at first Your investors should want the kind of return your business is designed to produce. Most VC funds need a large equity exit. A business that distributes profits every quarter may be attractive to its owners without ever producing that outcome. Angels and family offices can be a better fit. The structure also matters. Most VC funds cannot hold S-Corp shares, and many avoid LLCs because pass-through income creates tax complications for their investors. You can convert to a C-Corp if the business develops a durable software product or needs substantial outside capital. But waiting has a cost: time spent as a pass-through generally does not count toward the QSBS holding period. For a business primarily built to distribute cash, I think that can be a reasonable trade. Get advice on the conversion path before choosing your initial structure. ### 3\. Raise only what gets you to positive cash flow Estimate what it will cost to win customers, recruit experts and deliver work until customer payments cover your expenses. Add a buffer for delayed payments and lost accounts. Raise that amount. For many expert-services data companies, my estimate is $1M–$2M, and usually under $5M. The right number depends on payment terms, contractor commitments and how much infrastructure you need before the first delivery. Surge AI is a useful reference point: it reportedly exceeded $1B in annual revenue while bootstrapped and profitable. Once the business funds itself, another equity round should have a specific purpose. Identify what the capital will build, how it will increase future profits and why retained earnings cannot fund it. Fast revenue growth alone is not a reason to dilute the owners. ### 4\. Pay employees a share of the profit they produce For a business organized around expert delivery, I’d make cash compensation the main reward: a competitive base salary plus a substantial bonus tied to the team’s gross profit. That gives people a direct stake in work they can influence. A pod that wins a lab account, delivers excellent data and manages its costs should share in the profit it creates. The details determine whether this works: - **Define gross profit in writing.** Start with account revenue, then subtract contractor payments and the agreed delivery, QA and platform costs. People should understand the calculation before accepting the job. - **Pay quarterly.** Keep the connection between performance and compensation visible. - **Defer part of the bonus.** I’d hold back 25%–30% for 12 months to account for quality problems discovered after delivery. Define how refunds and rework affect that amount. - **Show candidates the economics.** Explain what different levels of team performance would mean for their pay. Use actual payout history when you have it. - **Keep Quality Control (QC) independent.** Teams rewarded for profit need a quality function that can block delivery without sacrificing its own compensation. Some candidates will prefer equity. That is a real recruiting tradeoff. But for people who want clear upside tied to the value they create, a transparent profit share can be compelling. I have met several Strategic Project Leads and Engagement Managers working at AI Data Companies who have made mid 7 figures in cash in a couple of years. That's compelling in and of itself, without any worries about whether company equity will ever be worth something. ### 5\. Pay taxes first, reinvest 30%, distribute the rest Give every dollar of cash a clear order of priority. - **Pay people.** Cover salaries, contractor payments, bonuses and payroll taxes. - **Make tax distributions to owners.** Pass-through owners can owe tax on profits even when the company retains the cash. Estimate that liability quarterly and distribute enough ahead of payment deadlines, using a tax-rate assumption agreed with your accountant. - **Maintain an operating reserve.** Keep enough cash to handle delayed collections, committed contractor payments and the loss of a major lab account. Set the reserve against those obligations. - **Reinvest 30% of the remaining cash.** That would be my starting point. Spend it on shared assets that improve delivery and margins: reusable datasets, task-design tools, automated quality checks, expert onboarding, payment systems and secure facilities. - **Distribute the rest.** Retaining more should require a clear use for the money and an expected return. Customer concentration makes the reserve essential. A lab can reduce spending or switch suppliers much faster than you can replace the revenue. Your ability to survive that loss should not depend on raising an emergency round. Write the distribution policy into the company’s governing documents. S-Corp distributions must be proportional to ownership, so the tax policy needs to work within that constraint. Ask your accountant whether a state pass-through entity tax election would also improve the economics. The discipline is simple: protect the business, fund improvements that make people more productive and pay owners the surplus. ## The org: run it like a multi-manager fund **(a) Small pods that own a P&L.** Set up one pod per lab or per vertical, such as coding, math, medicine or law. Each pod has a lead, a few operators and its own contractor pool. The lead owns the lab relationship, delivery and margin. The pod's bonus pool comes from the pod's gross profit. **(b) A thin shared platform.** This covers tooling for task design, contractor onboarding, payments, security and data delivery. Build it once and let every pod use it. The 30% reinvestment pays for most of it. Charge pods for it so the cost shows up in their P&L. Keep headcount here low. **(c) Independent QC with veto power.** This is the most important function in the company. A pod paid on gross profit has every reason to ship fast and cut corners. QC is the check on that. QC reports to the CEO. It can block any delivery. Pay QA staff on accepted quality and lab feedback, never on volume. One bad dataset can lose a lab account that took 2 years to win. **(d) A tiny full-time core.** This is finance, legal, expert recruiting, and security and compliance. Labs will ask for SOC 2 and strict data handling, so security belongs in the core. Keep the core to a handful of people. **(e) An elastic expert network.** These are PhDs, doctors, lawyers and senior engineers on contract. You add them when a project starts and release them when it ends. Most of your cost and most of your quality come from this group. Worker classification is a real legal risk, especially in California under AB5\. Get it reviewed early. Multi-manager funds also shut down pods that stop making money. Do the same. Say a pod's lab account shrinks for 2 straight quarters with no new work in sight. Move its people to another pod or let them go. When the edge decays, some pods should end. ## A recruiting firm already runs this way A partner at a top recruiting firm emailed me about this framework. He loved the hedge fund comparison, and said his firm already looks like the model above: - **S-Corp.** Profits pass through and are taxed once. - **Bootstrapped.** The firm has never raised outside money. - **Profitable.** No further clarification needed :) - **Tiered commissions tied to production.** Recruiters get paid on the placements they make. The rate climbs as they produce more. - **Seven figures plus invested in data, tech and engineering** to make recruiters more productive. Recruiting fits the same three tests as AI data companies. A firm has to continuously refresh its candidate pool; a placement fee is easy to trace to the recruiter who earned it; and the business is dependent on excellent people. ## When this is the wrong structure The structure above is the default for a data business that is mostly expert services. However, some data companies sell software to enterprises or AI labs; this is real technology such as an eval platform, an RL environment product. Those companies have real equity value, and should likely use a C-Corp structure. Similarly, another set of companies has a core business (eg: vertical software) generating "AI data exhaust" that they then sell to the labs, using the proceeds to lower the price for their core offering (thus creating a virtuous flywheel). These companies should also run as C-Corp since data is not their core business but a way to accelerate the core. **Bottom line:** Before incorporating your company, figure out where the lasting value will sit: in software and proprietary assets, or in teams that deliver excellent work. That answer should shape your capital, compensation and ownership. Enjoyed this piece? Please share it with a friend, and consider subscribing if you haven’t already. Subscribe Check your inbox to confirm your subscription.