Alexandr Wang

Halfway through his Y Combinator batch, Alexandr Wang abandoned the chatbot for doctors he had arrived with, a copy of what the founders around him were building, and looked instead at what all those chatbots actually needed: labeled data and human effort to check it. The domain scaleapi.com was available, and a week later he launched on Product Hunt under a tag that read like science fiction, "an API for human labor."¹

Wang is the co-founder of Scale AI, a company that sells labeled training data and evaluation infrastructure to companies building artificial intelligence systems. Scale does not build AI models; it supplies the human-annotated data model developers need to train and test them. Wang started the company at 19, ran it for nine years, and in 2025 left day-to-day leadership to become Meta's first chief AI officer, running the division Meta built to catch up in the race.

Background

Wang was born in January 1997 in Los Alamos, New Mexico, to Chinese immigrant parents who both worked as physicists at the national lab. Growing up in that small lab town, he gravitated early toward math and competitive programming: a USACO finalist (2012, 2013), Math Olympiad Program qualifier (2013), and US Physics Team member (2014) at Los Alamos High School.

As a teenager, Wang attended rationalist-community camps organized in part by Paul Christiano, who later invented the RLHF technique used to align large language models; Greg Brockman and Eliezer Yudkowsky also spoke there. It left him thinking AI might be "the most important thing to work on in my lifetime."¹ Before college he worked as a software engineer at Addepar, then at Quora and, as an intern, Hudson River Trading. He enrolled at MIT at 18, applied to Y Combinator soon after, and dropped out.

Getting started

Wang's original idea, developed with co-founder Lucy Guo, was a chatbot for doctors. He has since called it "mimetic," a copy of what other founders in the batch were building rather than anything he actually understood: by his own account, he knew nothing about doctors or medicine, and the batch left him "quite lost mid-batch."

The idea that stuck came from watching what other chatbot startups actually needed: labeled data and human effort to train and check it, poorly served by the incumbent, Amazon Mechanical Turk. Wang's insight was to stop building a chatbot and instead sell that data and labor. The domain scaleapi.com was available; he registered it and launched Scale on Product Hunt a week later, tagged "an API for human labor." An ex-YC founder at Cruise saw the post, and Cruise quickly became Scale's largest customer, pulling it into self-driving-car labeling as its first vertical.¹

What he built

Scale grew by repeatedly relocating to whichever part of the AI industry needed labeled data next: sensor data for autonomous vehicle companies first, then data for large language model developers as that field grew from the GPT-2 era onward, including 2019 work with OpenAI touching the RLHF work behind InstructGPT, a ChatGPT precursor. From there it expanded into enterprise AI and defense programs, including Thunder Forge, a DoD military-planning system Wang says compresses a roughly 72-hour decision cycle to about 10 minutes.¹

Funding tracked that growth: a $100 million Founders Fund investment made Scale a unicorn in 2019; a 2020 Tiger Global round valued it above $3.5 billion; a 2021 round of $325 million valued it near $7.3 billion, briefly making Wang, at 24, one of the youngest self-made billionaires; and a 2024 round of roughly $1 billion valued it at $14 billion. Scale also built Humanity's Last Exam, a benchmark meant to resist gaming; frontier scores on it rose from roughly 7 percent to above 20 percent within months.

In June 2025, Meta invested about $14.3 billion in Scale for a roughly 49 percent stake, valuing it near $29 billion. Wang stepped down as CEO to lead Meta's new Superintelligence Labs as its first chief AI officer, remaining on Scale's board.²

How he operates

Wang's hiring approach was hands-on to an unusual degree: he personally reviewed and approved or rejected every hire at Scale, regardless of company size, applying a principle he calls "quality is fractal," the idea that standards trickle down from leaders and erode once employees sense their manager doesn't care. He has said the most reliable predictor of success at Scale was not technical skill but how invested people were in their own work, a signal he summarized as looking for people who "give a damn."¹

Wang's broader pattern as a builder has been to place Scale ahead of each new wave of AI activity rather than commit to one bet, comparing it to Jensen Huang's approach at Nvidia: positioning the company to be ready before a trend materializes, since data demand precedes AI adoption.¹

Where things stand

As of 2026, Wang leads Meta Superintelligence Labs, having spent roughly a year rebuilding Meta's AI research organization after concluding its prior flagship, Llama 4, had fallen short. That rebuild produced Muse Spark, released in early 2026 and described by Wang as an "appetizer" rather than a frontier model, with a more capable "entree" flagship in training. Meta withheld its weights after early training triggered high-risk classifications around biological risk, opting to deploy it with mitigations.² Scale continues to operate separately, with Wang retaining a board seat.

Key facts

• Born January 1997 in Los Alamos, New Mexico, to Chinese immigrant parents, both physicists at the national lab.
• Co-founded Scale AI in 2016 with Lucy Guo via YC's Winter 2016 batch, at 19, after dropping out of MIT and abandoning a chatbot-for-doctors idea.
• Personally reviewed and approved or rejected every hire at Scale, per his "quality is fractal" philosophy.
• Scale's valuation rose from about $1 billion (2019) to $3.5 billion (2020) to $7.3 billion (2021) to $14 billion (2024) to roughly $29 billion (2025).
• In June 2025, Meta invested $14.3 billion in Scale for a roughly 49 percent stake and hired Wang as its first chief AI officer.
• Scale built Humanity's Last Exam, an AI benchmark, and Thunder Forge, a DoD military-planning system with Indo-Pacific Command.