The human data behind the frontier AI models
HQ: San Francisco, CA, United States | Founded: 2020 | Employees: ~100–130 full-time (founder cites under 100 at the $1B revenue mark) | Stage: Bootstrapped and profitable; reportedly preparing a first outside raise of up to $1B at a $15B+ valuation (2025) | Website: https://surgehq.ai
Surge AI does the human work behind artificial intelligence. Its network of workers and domain experts labels and annotates data, ranks and compares model outputs, writes preference data for reinforcement learning from human feedback (RLHF), red-teams systems for safety failures, and evaluates model answers in specialist fields like law, medicine and coding. For the handful of companies training frontier models, this work is not optional — a model is only as good as the human feedback it learns from.
The company was founded in 2020 by Edwin Chen, who had built machine-learning systems at Google, Meta and Twitter and believed the industry's data pipelines — typically run through crowdsourced gig platforms — produced sloppy work. Surge's model is different: carefully managed, well-paid workforces and expert evaluators, run as a service. It was profitable early and never needed investors.
By 2024 the quiet approach had produced an unusual outcome: more revenue than Scale AI, the venture-backed category leader, from a fraction of the staff. When Meta bought 49% of Scale in mid-2025 and took its CEO with it, labs looking for a neutral data partner had fewer places to turn — and Surge, still independent, was suddenly one of the most strategically important companies in AI.
Surge AI is the most profitable company in AI that nobody had heard of until 2025. Edwin Chen — a machine-learning engineer out of Google, Meta and Twitter — founded it in 2020 to do the unglamorous work of AI: paying and managing skilled humans to label data, rank model outputs, red-team systems and evaluate answers in expert domains. Every frontier model depends on this work; Surge just did it better and quietly took the market.
The numbers are the story. The company passed $1 billion in revenue in under four years with fewer than 100 employees, completely bootstrapped — what Lenny Rachitsky called the fastest company in history to the mark. Forbes put 2024 revenue at $1.2 billion, ahead of Scale AI's reported $870 million, and third-party estimates put 2025 revenue around $1.4 billion. No board, no venture dilution; Chen owns the company.
Two market shifts worked in Surge's favour. First, model quality stopped being about raw data volume and started being about high-quality human feedback and expert evaluation — Surge's core business. Second, in June 2025 Meta paid $14.3 billion for 49% of rival Scale AI and took its CEO, and labs that had treated Scale as neutral infrastructure suddenly needed alternatives. Surge was the obvious one.
The risks are concentration and opacity. Revenue depends on a handful of frontier labs whose internal strategies can change overnight, the workforces behind 'human data' raise well-documented labour and ethical questions, and the company reports nothing publicly — almost every number here is third-party reporting. In 2025 Reuters reported it hired advisors to raise as much as $1 billion in its first outside capital, at a reported $15B+ valuation; later reports floated figures up to $25 billion. If that round closes, it converts a bootstrapped oddity into a fully-priced AI infrastructure giant.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki