Handshake training revenue surge

Handshake’s AI training business climbed to nearly $1 billion in annualized gross revenue, up from about $550 million in January, according to The Information. (theinformation.com) The company started 12 years ago as a job site for college graduates. It launched its human data-labeling business about a year ago, and TechCrunch reported in January that Handshake was forecast to end 2025 at a $300 million annualized revenue run rate and was on track for “high hundreds of millions” in 2026. (techcrunch.com) Handshake says its newer unit sells “expert human model validation and technology systems” to frontier AI developers. The work includes writing domain-specific prompts, grading model answers, annotating datasets, and suggesting improvements to outputs. (joinhandshake.com, support.joinhandshake.com) The labor pool is a core part of the pitch. Handshake says it works through 1,500 university partners in the United States and Europe and has a network of 18 million students and alumni, including 3 million graduate-level scholars and more than 500,000 Doctor of Philosophy degree holders. (joinhandshake.com) That network lets Handshake argue it is closer to the source of expert labor than many rivals. In January, Cleanlab co-founder Curtis Northcutt told TechCrunch that data-labeling competitors including Mercor and Scale AI often use Handshake’s platform to source doctors, lawyers, and scientists for projects. (techcrunch.com) The market around that work has grown fast as model makers chase more specialized human feedback. The Information reported on April 13 that AI firms are paying lawyers, Doctor of Philosophy degree holders, and medical doctors to grade model answers, and TechCrunch reported this month that Mercor had raised a $350 million Series C at a $10 billion valuation. (theinformation.com, techcrunch.com) Handshake has also been building tools to tighten quality control around that labor. In January it acquired Cleanlab, a startup whose software flags incorrect labels without sending every task to a second human reviewer. (techcrunch.com) The company says projects are flexible and part-time, with no minimum hours and a typical contribution of 10 to 20 hours a week. Pay varies by assignment, and Handshake advertises rates of up to $100 an hour for some fellowship work. (support.joinhandshake.com, joinhandshake.com) The rush into expert labeling has also sharpened questions about concentration and risk in the AI supply chain. TechCrunch reported this month that Mercor disclosed a breach tied to the open-source tool LiteLLM, and said Meta had paused contracts with Mercor indefinitely, citing Wired. (techcrunch.com) For Handshake, the shift is visible in the business mix. A company built for campus recruiting is now selling access to human judgment at scale, and the revenue figures reported this week show how much AI labs are paying for that work. (joinhandshake.com, theinformation.com)