How Young Billionaires Got Rich Off AI: Building Empires in the 18 Percent Window

According to a late-2025 Federal Reserve reading, only 18 percent of American firms have truly adopted AI. The video flips that stat into opportunity: if 82 percent are still on the sidelines, the race has just started. It tells the story of six young billionaires who did not invent a chatbot; instead they sold the missing piece — humans who can tell where a model goes wrong — and scaled that judgment.

The Billion-Dollar Idea: The Lab's Human Problem

Mercor began as a boring recruiting operation built at a São Paulo hackathon to match overseas engineers with American companies, handling paperwork and payments for a small cut. Founders Brendan Foody, Adarsh Hiremath and Surya Midha — three friends from the Bellarmine College Preparatory debate team — heard the same line twice in meetings with OpenAI and xAI: compute and text are abundant, but people who can judge whether an answer is good are scarce. In 2023 they moved the engine from the dorm, and with a Thiel Fellowship in 2024 they left college to chase that scarcity. The pivot was simple: point the recruiting machine at doctors, lawyers, bankers and engineers who could grade model output.

The model is lean and paid by the hour. Mercor hires working professionals at 75 to 150 dollars per hour, they mark a model's answer and write the scoring guide the lab then uses for the next ten thousand responses. By October 2025 the firm managed about 30,000 contractors and paid roughly 2 million dollars per day. Customers included OpenAI and Anthropic. That month it hit a 10 billion dollar valuation; owning about 22 percent each, the three 22-year-olds became the world's youngest self-made billionaires at about 2.2 billion dollars apiece, a year younger than Zuckerberg at 23, two years after a hackathon with no degree and no patent.

To a Billion Alone: The Bootstrapped Lesson

Edwin Chen's Surge AI answers the same bottleneck with the opposite capital story. The son of Taiwanese immigrants who ran a restaurant, he worked the family kitchen as a teen, studied math, computer science and linguistics at MIT, then spent a decade as a research scientist at Google, Facebook, Twitter and the hedge fund Clarium. Disliking the Valley status game, he started the company alone in 2020 from an apartment with saved earnings — no co-founder, no deck, no outside money. In 2024 with about 110 people it reached 1.2 billion dollars in revenue, serving Google and Anthropic. Owning roughly three quarters, Chen became the youngest member of the Forbes 400 and the fastest ever to a billion in revenue without outside funding. The takeaway is order: sell work you can already do to a customer who has paid before, then scale.

Take the Job Nobody Wants

Scale AI grew by taking the job everyone needed and no one wanted. Alexandr Wang grew up in Los Alamos, the town built for the atomic bomb, to two weapons physicists who had come from China. He chased a Disney trip prize in sixth grade math, then made the US math olympiad program, the physics team and two national computing finals. While trying to automate his own fridge he saw the blockage was not algorithms but unlabeled data. He left MIT after a year at 19. A failed 2015 appointment-booking app with designer Lucy Guo — they spent their days calling doctors instead of building — turned into a joke that became a company: an API for humans. Real people drew boxes around every car, stop sign and pedestrian, millions at a time, so models could learn what they see. In June 2025 Meta paid 14.3 billion dollars for 49 percent, valuing Scale above 29 billion; Wang, 29, joined Meta as chief AI officer with wealth around 3.6 billion. The lesson is to package the apology task — the thing a client says sorry to ask for — as a monthly service.

Lucy Guo is the other half of that story. Studying computer science at Carnegie Mellon, she took the Thiel Fellowship's 100,000 dollars to leave school, became the first woman designer at Snapchat, worked at Quora and met Wang. At 21 they co-founded Scale in 2016; she left in 2018 after disagreements. For seven years she had no title and no say, but kept her shares. When Meta's deal closed in 2025, Forbes named her at 30 the world's youngest self-made woman billionaire with about 1.25 billion dollars. In December 2022 she launched Passes, where creators sell directly to followers, raising about 50 million dollars with names like Olivia Dunne and Bella Thorne. The point is to start the next thing before you know how the last one ends, with two hours on a weekend.

Build for the Job You Do: From Desert to Code

Michael Truell and Anyphere's Cursor is a story of a wasted year and a return to the familiar. Truell wrote code at 11 for his own mobile games, built a programming game at 14, made the national computing finals and co-created the 10,000-entrant Highlight contest at MIT. In 2022 he and three MIT classmates — including Aman Sanger from New York — started Anyphere. They chased mechanical engineering tools for physical parts, a field they did not know, which Truell calls wandering in the desert. They pivoted to the job they lived every day: writing code with slow tools. Cursor charged 20 to 40 dollars a month while paying more to OpenAI and Anthropic to run it, leaving about a 30 percent gross margin. In April 2025 its support agent hallucinated a login rule and triggered a cancellation wave before a human could step in. In November 2025 it crossed 1 billion dollars in annual revenue at a 29.3 billion valuation; on August 14, 2026 SpaceX bought it for 60 billion in stock, the largest startup acquisition on record, with about 2.4 billion per founder in rocket-company stock.

Aravind Srinivas built Perplexity by attacking search where the incumbent cannot follow without hurting itself. Raised in middle-class Chennai near the IIT Madras campus his mother pointed at, he missed a computer science seat by a fraction, fell into depression, taught himself Python at night, competed on Kaggle and graduated top of his dual degree in 2017. A Berkeley PhD with Peter Abbeel, a DeepMind internship in London where he slept at the office due to bad housing, a book on how Google was built, and a humbling stint at OpenAI in 2018 shaped him. He returned full-time at 27 in 2021 and quit in August 2022, weeks before ChatGPT, when almost no one outside labs saw it coming. With Dennis Yarats and Andy Konwinski he started Perplexity; the first product searched Twitter in English, then the citation-backed answer engine came, with Jeff Bezos backing. In 2025 it reached a 20 billion valuation with 100 million monthly users; Srinivas, 32, was worth about 2.5 billion. Google made 63 billion from search in the final quarter of 2025, all advertising — a page that keeps you stays paid, an answer that sends you onward does not.

The One-Person Shop Model

The second act moves to Tony Dinh and TypingMind, the live proof of one person, one problem. He left a safe six-figure big-tech role and shipped tiny products in public until the AI wave. His cleaner, faster chat interface with extras for power users now earns over 130,000 dollars a month — at times 145,000 in monthly recurring revenue — posted publicly, with no employees, no office, just a laptop while traveling. No investors, no fifty hires, just one expensive problem solved with AI doing the work of a team.

Step one is to avoid being a generalist. The mistake is offering a little of everything; money rewards specificity. The test given is to look for what businesses already pay people to do that is boring, repetitive and time-heavy: writing product copy, answering the same questions, cutting video, building reports, chasing leads. Each already has a budget; each can be handled in a fraction of the time once you narrow to one person and one headache and state it in one sentence.

Step two is to start high, not cheap. A 20-dollar ebook or 50-dollar mini-course needs thousands of customers and a heavy marketing engine you do not have. The math flips at 2,000 dollars a month: four clients yield 96,000 a year, eight approach 200,000, managed from a kitchen table. High tickets also signal seriousness — clients show up, get results and refer — and let you deliver a memorable experience to a few names you actually know.

Step three is to let the system carry the repetitive load. Filter everything through eliminate, automate, delegate — in that order. Cut what no one reads, like meeting notes no one opened. Automate what repeats: turning one long video into 20 clips, transcribing calls, drafting outreach and first drafts, organizing research. Delegate only what the system should not do, starting with five to ten hours a week for a part-time helper and learning to hand work off. The example given is 30 posts, a weekly newsletter and inbox coverage — three hires a few years ago, now an afternoon of first drafts for a few dollars plus human taste.

Step four is to build relationships, not an audience. Chasing virality for a year often ends broke. The alternative is a list of 50 connectors — each could be worth six or seven figures — not customers but people who already have the network you want. Give first: invite them to a simple show, make them look good for 45 minutes, build a real bond. The video cites clients who built seven-figure businesses with fewer than 50 targeted subscribers and a coach who moved from two to fifty-five high-ticket enrollees in two weeks via free live coaching that led with value.

Steps five and six close the loop: stay lean on purpose and start before ready. The goal is not the biggest headcount but the most freedom and retained cash; without payroll and rent a slow month is survivable and risk is affordable. Lean is the advantage because software covers the team. Tony shipped imperfect products in public and learned in the open. The assignment is one problem, one offer and one real conversation this week — no website, logo or business plan — because tools more powerful than what a billion-dollar firm had a few years ago are now nearly free.

Orders Are Not Money: The Cash-Flow Lesson

The final act teaches money. In 1992 Damon John had 40 dollars in Queens. His mother taught him to sew; he bought fabric, made a few dozen hats, sold them on a street corner for 800 dollars in a day, then expanded to shirts and jerseys at a flea market. His mother mortgaged their house for 100,000 dollars and turned half of it into a factory. At the Magic apparel show in Las Vegas he wrote 300,000 dollars in orders — but stores pay 30, 60, 90 days after delivery while fabric and sewing are paid up front. The house money covered about 75,000 in production and was gone in four months with no inflow. Twenty-seven banks said no while orders swelled to about one million. An ad in the paper drew 33 calls; Samsung's textile arm funded production on condition of 5 million in sales within three years. He did 30 million in three months and by 1998 did over 350 million worldwide. The line is simple: orders are not money.

The same lesson is sized to a small firm via a dance studio that looked profitable for the year but ran October -5,300, November +2,500, December -900, down 3,700 for the quarter. Fixed costs were about 20,000 a month — 11,000 rent, 2,000 insurance plus instructors, software and cleaning. Knowing the floor turns guessing into a target: 25,000 to keep something left. Your floor is the sum of what must go out every month; divide cash on hand by that floor to get runway. With six months you can walk from a bad client; with three weeks you say yes to everything that makes the next three weeks worse.

Price and the Quiet Power of Compounding

Revenue problems are often pricing problems. The anchoring example states the mastermind at 50,000 a year before a 15,000 four-month program, so 15 sounds reasonable and smaller entries sound like bargains; starting low anchors low forever. Raises apply to the next new client, avoiding awkward talks, and you sell outcomes not hours — adding 200,000 in revenue is worth ten times six sessions. Low tickets at 50 a month demand thousands of customers, marketing and support; high tickets need two wins at 40,000 to 100,000 a year, or one platform deal at a million.

How you pay yourself matters. In the Canadian example money left in the corporation is taxed around 19 to 20 percent while a large personal salary can approach 50; pulling everything at once is costly. He pays a small steady salary for a clean personal record and keeps the rest working. If you work from home and the office is one of eight rooms, about 12 percent of mortgage interest, utilities and internet can be a business cost, receipts kept and checked with an accountant — rules differ by country. Growth should then be boring: 10,000 at 5 percent becomes about 16,000 after ten years and 43,000 after thirty; the first decade earns 6,000, the last twenty earn 27,000, so waiting costs the years where compounding does almost everything. Put it in a broad index, leave it, and avoid stock picking that steals hours from the work that actually moves income; cash is not safe either, with prices up about 22.5 percent since 2020. And compounding works against you at around 20 percent on credit cards — no reliable investment beats clearing that balance first.

The through-line is to look at three numbers for 20 minutes a week — what came in, what went out and what remains — until it stops feeling scary. The 27 percent stat on basic money questions is not about being a numbers person but about never being taught; deciding you are someone who reads statements turns avoidance into habit. Start with ten names for the list of fifty tonight; that list costs an afternoon and becomes the highest-value page in the business.