Micro1 Hits $500M Run Rate as It Chases Mercor and Handshake

Micro1, a four-year-old startup that pays doctors, lawyers, engineers and other specialists to help train AI models, went from a $100 million gross annual run rate to $500 million in eight months, according to TechCrunch, which cited a person familiar with the company. The startup keeps roughly 60% to 70% of that figure after paying contractors, putting net run rate somewhere between $150 million and $200 million. It's still well behind the two biggest names in the space: Mercor hit $2 billion in gross annualized revenue this summer, and Handshake reached $1 billion earlier in the year. But the pace of Micro1's climb is the part getting attention. Inc. reported separately that the company's annual recurring revenue went from $7 million in January 2025 to over $400 million by the time it profiled founder Ali Ansari in August 2026. Micro1 raised $35 million at a $500 million valuation last September; TechCrunch reports the company may have since closed another round at a valuation well above that, and Inc. puts the current figure above $2.5 billion.

The base business is what it sounds like: AI labs and corporations pay Micro1 to have contract experts label data and evaluate model outputs, a process often called reinforcement learning from human feedback. Contractors on the platform report pay between $50 and $200 an hour, with some going higher for specialized skills. But Micro1 is pushing to generate more of its data without a human in the loop every time, and to resell the same dataset to more than one client. Automated video-description generation is one example the company points to. A person familiar with Micro1's finances told TechCrunch that this "off-the-shelf" data, sold to multiple customers, carries gross margins of 80% to 90%, well above what one-off, human-labeled work typically returns.

Ansari denies selling data to Chinese labs

Selling the same dataset more than once is also where the controversy sits. Critics argue that data-labeling firms reselling off-the-shelf datasets to Chinese AI developers help close the capability gap with US labs. Ansari has said publicly that Micro1 doesn't do this. Posting on X last month, he wrote: "Some human data companies work with foreign adversaries. And the results show today in Kimi K3. We believe it's shameful to claim American AI dominance desires while selling millions worth of data to countries that we are in adversarial competition with."

Micro1 didn't start as a data company. Ansari built it at Berkeley as an AI recruiting agent, using an AI screener called Zara to vet software engineers before he'd spend time interviewing them himself. The pivot happened almost by accident: one of Micro1's data-labeling clients asked the company to source 600 engineers in three weeks, and Ansari realized the client's actual bottleneck wasn't recruiting software, it was finding enough vetted experts. In early 2025, while starting a master's at Stanford, he repositioned Micro1 as a data-labeling and AI training infrastructure company and landed its first frontier-lab client. Zara is still doing the sourcing work underneath the data business: Ansari told Inc. that a few thousand candidates go through its interview process daily, with roughly 20% accepted, and the company has logged more than five million signups and around three million interviews since Zara launched.

Ansari is 25. His family immigrated from Tehran to Los Angeles in 2010 after winning the US green-card lottery; his father restarted a kitchen-remodeling business and his mother opened a daycare, both still running. Ansari founded Micro1 as a Berkeley undergrad in 2022, then applied to Stanford's master's program, he told Inc., largely to get access to AI research circles. He spent a year cold-messaging Stanford professor Andrew Maas before Maas agreed to join as Micro1's VP of engineering, employee number 80. Ansari is now on leave from Stanford to run the company full time.

Data labeling for AI has gone from a niche outsourcing category to one of the more contested layers of the AI economy. Scale AI, founded in 2016, built its business on annotation work before Meta acquired 49% of it for $14.3 billion in mid-2025, a deal that installed Scale founder Alexandr Wang as chief AI officer of Meta's Superintelligence Labs and pushed several AI labs to look for alternative vendors. Surge AI, founded in 2020, picked up much of that overflow and reportedly crossed $1 billion in revenue by 2024; it's now reportedly seeking its first funding round in five years at a $25 billion valuation. Mercor, founded in 2023 by three then-19-year-old friends, and Handshake, a decade-old recruiting platform that entered data labeling in 2024, round out the group of expert-focused competitors Micro1 is up against. Mercor co-founder Adarsh Hiremath told Forbes after the Scale-Meta deal: "It just doesn't happen too often in startups where your biggest competitor gets torpedoed overnight." Reports have also circulated that Mercor tried to poach Micro1 staff with cash bonuses as the two companies compete for researchers and salespeople.

Micro1 is also chasing data outside the labeling business entirely. When Spirit Airlines went through bankruptcy, Micro1 reportedly offered $12.5 million for the carrier's corporate data, outbidding a $10 million offer from Google. The interest reflects a broader shift: with public internet text increasingly exhausted as a training source, AI labs are paying for proprietary datasets that can't be scraped, and ordinary companies are starting to treat years of internal records as a sellable asset.

Ansari is also betting the company's next act won't be data labeling at all. Cortex, Micro1's AI agent evaluation product, targets enterprises building or deploying their own agents and smaller models rather than frontier labs. "There's going to be a million times more agents built than models," Ansari told Inc., predicting Cortex will overtake the frontier-lab training business within two to four years. That shift, along with the company's push toward robotics pre-training, having generalist contractors record everyday object interactions in their own homes, suggests the highest-margin work increasingly involves less repeat human labor per dataset, not more; specialist evaluation work, the kind Zara screens candidates for, looks like the more durable lane as the company leans harder into scale.

Update, September 28, 2026: Ansari put a number on the robotics push in a LinkedIn post this week, writing that Micro1's robotics department grew from zero to $100 million in annual recurring revenue in nine months. That would be roughly a fifth of the $500 million gross run rate above, though he didn't say whether his figure is gross or net, and no outlet has confirmed it yet. He also announced Micro1's first robotics data lab, in Malibu, focused on evaluating models on physical hardware, and said the company is hiring robotics researchers, engineers and teleoperators for it. Some of that on-site work is already listed in the Los Angeles area, including Robot Wrangler and Hardware Tester contract roles. The rest of the openings are on the Micro1 jobs page.