The $200 million middleman between humans and machines

A pair of men's loafers sold for $50 on eBay in 2013 launched what may become the fastest path to a billion-dollar fortune in the current AI cycle. The seller was twelve years old.

Ali Ansari did not know he was training for the AI economy when he biked through Woodland Hills, buying whatever he could carry from garage sales and thrift stores to resell online. Nobody told him that a textbook arbitrage business clearing $100,000 before his sixteenth birthday, buying used books from college students through a website he coded himself and flipping them on Amazon at a 50% markup, was preparation for anything. Same with the software agency at UC Berkeley, building websites for small businesses, where he discovered that recruiting engineers consumed more time than the engineering itself.

What Ansari did know, before most venture capitalists caught on, was that finding qualified people fast and at scale was a business worth automating. The AI screening tool he built to hire developers for his agency became Micro1. Three years later, the company generates $200 million in annualized revenue, and Forbes estimates that ongoing funding conversations value it at $2.5 billion. Ansari is 25.

The accidental supply chain

Micro1 exists because AI models eat human knowledge for breakfast, and nobody has figured out how to fake the meal.

GPT, Claude, Gemini, pick your favorite. They all get smarter the same way, and it is not elegant. Reinforcement learning from human feedback means real people sit with model outputs and mark every mistake. This answer beats that one. The code breaks right here. That legal argument? Garbage after paragraph two, and let me show you why. The work is tedious. It costs a fortune. And it swallows people whole. One frontier model needs thousands of experts grinding corrections around the clock. Coders, doctors, lawyers, and, because someone has to, comedians.

Micro1 finds those people. Zara, the company's AI recruiter, does the first pass, throwing coding problems and domain quizzes and behavioral questions at anyone who applies. Something like 130,000 candidates have sat through it. A hundred specialties, 60 languages. Pass Zara's filter and you land in a talent pool that gets deployed to AI labs and Fortune 100 clients. Microsoft is one of them. Most of the others hide behind NDAs.

Strip away the AI branding and the business model is old-fashioned. Micro1 takes a margin on the labor it supplies. Somewhere in Bangalore, a PhD mathematician sits in front of a laptop rating whether a language model solved an integral correctly. Micro1 found that mathematician, tested her, and bills the client for every hour she spends training, testing, and correcting AI systems. Ansari's company does not own the data. It does not train models. It sells access to the people who make other companies' models better.

"We are the AI platform for human intelligence," Ansari told the Stanford Daily in October 2025. The slogan captures the contradiction cleanly. The most capital-intensive technology sector in history depends on an army of freelance humans whose skills cannot yet be replaced by the systems they are building.

$4 million to $200 million in fourteen months

Revenue like this makes venture investors lose discipline.

In 2024, Micro1 was doing $4 million a year. By January 2025, $7 million. Then the curve went vertical. September: $50 million, on the back of a $35 million Series A at a $500 million valuation. December: $100 million. And by early 2026, Ansari told the LA Times the figure had crossed $200 million. That is a 50x jump in fourteen months. Read that number again.

01 Advisors led the Series A. Dick Costolo and Adam Bain, the former CEO and COO of Twitter, co-founded the firm. Bain took a board seat at Micro1. So did Joshua Browder, who built DoNotPay. Total money in: $41.6 million across four rounds going back to a $3.3 million pre-seed in August 2023 from Companyon Ventures.

Impressive, until you look sideways. Mercor, founded the same year, closed a $350 million Series C last October and is now worth $10 billion on $450 million in annual revenue. Surge AI is even wilder. Bootstrapped by one person, no outside money, $1.4 billion in revenue with a 121-person team, and last summer it was fielding calls about its first fundraise at a reported $25 billion valuation. Then there is Scale AI, which was the dominant player before Meta acquired 49% of the company for $14.3 billion, swallowing $870 million in annual revenue whole.

Micro1 is the smallest company in a market where the smallest company still grows at speeds that would make a SaaS founder weep.

The earthquake that created the opening

Micro1's growth story is inseparable from a single event: Meta's June 2025 acquisition of a 49% stake in Scale AI for $14.3 billion.

The deal made Alexandr Wang, Scale's 27-year-old founder, one of the wealthiest people in technology. It also detonated his client base.

Google, which had planned to spend approximately $200 million with Scale in 2025 for human-annotated data to train Gemini, began winding down the relationship within weeks. Nervous Google executives saw a nightmare scenario: proprietary training methodologies and data leaking to Meta through Scale's new ownership structure. OpenAI followed, cutting ties days after the announcement. The logic was identical and the anxiety was mutual: no frontier lab wanted its competitor's majority shareholder sitting inside its data supply chain.

Ryan Kolln, who runs Appen, had a front-row seat. "It was the equivalent of an oil pipeline exploding between Russia and Europe," he said. Demand at Handshake "tripled overnight," according to CEO Garrett Lord. Over at Turing, CEO Jonathan Siddharth watched $50 million in potential contracts materialize in two weeks. Everyone wanted a new supplier. Everyone wanted one yesterday.

Micro1 caught the wave. The company's revenue doubled between September and December 2025, the exact months when Google and OpenAI were scrambling for Scale AI replacements. Ansari has not publicly attributed his growth to the Scale disruption. He does not need to. The timing speaks clearly enough.

Zara does the hiring. Humans do the work.

Forget data labeling. What Micro1 actually sells is speed in hiring.

Zara runs the show. Candidates land on Micro1's website and the AI agent takes over, pushing them through domain tests, live coding problems, behavioral screens. No human recruiter touches the process until Zara decides someone is worth a closer look. People who clear the bar enter a talent pool that Micro1 can activate for client projects in days, not months.

Ansari had lived this problem before, at his Berkeley software agency. Hiring engineers to build websites consumed 80% of his time. Automation cut that to nearly zero. When AI labs started spending billions on human feedback data, the same bottleneck appeared at industrial scale: finding enough qualified people, fast enough, to meet demand.

Micro1's advantage is velocity. The company claims to have more than 130,000 deeply vetted candidates ready to deploy across 100 domains. Scale AI built its workforce over nearly a decade. Micro1 built its pipeline in three years by letting an AI agent do the initial screening and letting human quality reviews validate the results.

But here is a risk that should make anyone paying attention uneasy, and it is the same risk facing every staffing company in history. Micro1 does not own the labor. Contractors can leave. Clients can poach them. And competitors with deeper pockets, Mercor has raised $519 million, Surge AI brought in $1.4 billion in revenue with no outside funding, can outbid Micro1 for the same talent pool. Ansari himself acknowledges the pressure. Running a company in this market, he told the LA Times, "feels like being in a constant battle trying to meet demand, raise money, and punch back against competitors trying to poach his employees."

The synthetic data question

Competitors can be outrun. What keeps the executives at companies like Micro1 nervous is something harder to outpace: the possibility that AI models learn to train themselves.

Synthetic data, output generated by AI models and used to train other AI models, is the industry's most aggressive area of cost reduction. If frontier labs can replace expensive human feedback with cheaper machine-generated alternatives, the entire human data market contracts. Every company that brokers access to human experts becomes less necessary. And that $1 trillion market Ansari projects? It shrinks before it materializes.

Here is the thing, though. Models trained predominantly on synthetic data suffer from a phenomenon researchers call model collapse. A degradation loop where each generation of AI-generated training data introduces compounding errors that erode output quality. A species that breeds only with itself degenerates. The AI industry has bumped into this wall enough times to be cautious about tearing down the human scaffolding.

For high-stakes domains, medical reasoning, legal analysis, advanced mathematics, human experts remain the only reliable source of ground truth. If you are building a model that gives medical advice, you need a doctor to check its work. No shortcut exists for that. RLHF tasks like open-ended comparative ranking, safety identification, and contradiction detection command premium rates, sometimes exceeding $100 per comparison in late 2025. The experts who can provide this feedback are the same doctors, researchers, and specialists who are already scarce outside the AI training market.

Micro1's bet is that demand for this expertise accelerates rather than fades. The company recently expanded into robotics training data, recruiting approximately 1,000 people across 60 countries to record footage of themselves performing household tasks. Workers receive Ray-Ban smart glasses and capture themselves cooking, cleaning, and manipulating everyday objects. The footage trains robotic systems to operate in domestic environments.

"In the long run, human data will become a $1-trillion market," Ansari has said. He derives the projection from an assumption that roughly 5% of all human labor will eventually be redirected toward training AI systems. The math is speculative. The direction of travel is not.

The immigrant founder and the incentive structure

Ansari's biography reads like a Silicon Valley creation myth, and that is precisely what makes it useful for investors.

His family won the U.S. green card lottery and left Iran when he was ten. Back home, his father ran a kitchen cabinet factory in a beach town. In Woodland Hills, they crammed four people into one room at a relative's place. His mother pulled shifts at Target. Ansari, who could barely order lunch in English, spent more time in the principal's office than in class. Teachers kept calling home. "'Hey, your son's making cow noises again,'" he told the LA Times, laughing about it now.

All of the early hustle was real. EBay reselling at twelve. Textbook arbitrage at fourteen. A tutoring business in high school that he later sold. A software agency at UC Berkeley that generated enough revenue to fund Micro1's origins. He graduated in 2022 with a degree in math and computer science and incorporated Micro1 the same year.

None of this is accidental storytelling. Micro1 is approaching a fundraise that would value it at $2.5 billion, and Ansari needs investors to buy both the market and the man. Immigrant kid, garage sales, self-taught coder, hundred-grand textbook business before he could drive. That story converts skeptics in a pitch meeting. Try asking a venture capitalist for billions when you are 25 and your resume is thin. Now try it when you have been running businesses since middle school. Different conversation entirely.

Fame is new for Ansari, and he is still figuring it out. He hired a chief of staff recently, someone to handle the logistics of being a minor celebrity and a son at the same time. He bought his parents a house in Woodland Hills. Television bores him. Podcasts about business and biographies of founders, that is his idea of relaxation. Hobbies? "Work," he says, and he is not being coy about it.

"I feel very grateful and very stressed," he told the LA Times. "That kind of summarizes it."

What the middleman reveals about the market

Micro1's rise exposes a structural feature of the AI economy that the headlines about trillion-dollar model training budgets tend to obscure.

Consider what this means. The most valuable technology being built today depends on a workforce that is, by design, temporary. Human feedback trainers are not employees. They are contractors, sourced through platforms like Micro1, Mercor, and Surge, paid by the task, and deployed for as long as the current training run requires their expertise. When the run ends, the contract ends. When a new model needs different skills, the platform recruits different people.

Each leading AI company spends approximately $1 billion annually on human data, according to Time magazine. Those budgets are growing, not shrinking. The race for AI supremacy is, at the operational level, a labor procurement contest. Whoever can find the most qualified experts, screen them fastest, and deploy them at global scale has an advantage that raw compute cannot substitute.

Micro1 is a middleman. But in a market where the product is human intelligence, the middleman sits at the chokepoint.

The test

Twelve months from now, we will know whether Micro1 is a billion-dollar company or a fast-growing staffing firm that caught a wave.

The $2.5 billion valuation being discussed in current fundraising implies a revenue multiple that only makes sense if growth continues at its current trajectory and the market for human training data does not contract. Two conditions must hold. First, synthetic data must remain insufficient for the highest-value training tasks. Second, the Scale AI disruption must produce lasting market share shifts rather than a temporary scramble.

Ansari's father, whose loafers started all of this, recently told his son he should diversify into robots. When Ansari said Micro1 had already started, his father complained. "You stole my idea. You got to give me equity."

The old man has good instincts. Whether the company's valuation holds will depend on whether Ansari's instincts are just as sharp, not at a garage sale in Woodland Hills, but at the center of an industry spending more on human intelligence than any in history.