AI Momentum 107.6 +2.7%GitHub Pulse 5172 -5.9%Capital Flow $204.6B +5804.2%Hiring Heatmap 0.0 -100.0%
Advertisement
Talent & Organizations

Meta loses a star AI researcher to Mercor, a training-data startup valued at $20 billion

Meta has lost another senior researcher from the AI lab it staffed with some of the richest pay offers in tech. Jason Wei said on November 17 that he is leaving Meta's TBD Lab after about a year to become chief scientist at Mercor.

Mercor's main business is not building AI models. The San Francisco startup pays doctors, lawyers, bankers and other specialists to create and grade the material AI companies use to train and test their systems. Investors valued it at $20 billion just over a week ago.

That makes the move a signal worth reading. One of the best-known researchers in the field has chosen a supplier over a frontier lab. Wei wrote that as more companies train AI models, "data will shape the next frontier." If he is right, the firms that sell expert human judgment gain leverage alongside chipmakers and model builders.

Who is moving, and what he will do

Wei is best known as lead author of a 2022 Google paper on "chain-of-thought" prompting. The technique gets AI models to work through a problem step by step. It helped lay the groundwork for the reasoning features most major AI labs now sell.

He moved to OpenAI in 2023, where he worked on its reasoning models. He also built widely used tests that check whether AI answers are factual and whether AI can research the web or handle health questions. Meta hired him from OpenAI in July 2025, during Mark Zuckerberg's spending spree to staff a new superintelligence unit. Wired first reported that hire.

Mercor co-founder and CEO Brendan Foody said Wei will help build Mercor's research team and lead its data and evaluation work. Foody called Wei one of Mercor's favorite customers and thought partners in recent years. Neither man gave a start date, a team size or a reason for leaving. Wei called his year at Meta amazing.

Pay terms were not disclosed on either side. None of the reports reviewed included a comment from Meta.

Meta's retention problem

Wei is not the first prized hire to leave Meta's lab quickly. Ruoming Pang, Apple's former head of AI models, left Meta for OpenAI in February 2026 after about seven months. Reuters, citing The Information, reported the move. It also cited Bloomberg's earlier report that Pang's Meta package was worth more than $200 million over several years.

Chinese tech outlet QbitAI, citing commentators, pointed to a "one-year cliff." Large hiring packages often release a first big block of stock after twelve months, which makes leaving cheaper afterward. That is commentary, not a confirmed term of Wei's contract. He joined Meta in July 2025, so he is past that point.

For Meta shareholders, the question is what the spending is buying. In June 2025 Meta paid $14.3 billion for 49% of Scale AI, a leading company in Mercor's line of business. It also brought in Scale's chief executive, Alexandr Wang, for its superintelligence effort. Now one of Meta's marquee researchers is going to work for a Scale competitor.

What Mercor is building

Mercor announced a $750 million funding round on November 10 at a $20 billion valuation. That is double the $10 billion it was valued at in October 2025. Backers include Benchmark, General Catalyst, Nvidia, Salesforce and Zoom. The round also let employees sell some shares. Mercor did not say how much of the $750 million went to them rather than to the company.

The company says more than five million experts are in its network. It has not disclosed revenue. Tech Funding News reports that Mercor's yearly run rate passed $2 billion in July, but notes that figure is gross billings. It estimates contractors keep roughly 60% to 70% of it.

Mercor's pitch to investors is that every business deploying AI agents, software that carries out tasks on its own, will need custom tests built from its own work. Foody has said every job will need its own agents. The company already publishes a set of tests, called APEX, that grade AI on real professional work.

Wei is the latest in a run of research hires. In October Mercor launched a research team that includes Edward Hu, a former OpenAI researcher behind a widely used method for cheaply customizing AI models. Mercor said that team also has unnamed researchers from Anthropic and from hedge funds Bridgewater and D. E. Shaw.

The bet carries risk. Mercor built its business selling to frontier AI labs and is only now pushing into ordinary companies. By Tech Funding News's estimate, most of its billing also passes straight through to contractors. A $20 billion valuation assumes that enterprise demand arrives at scale.

What to watch: whether other senior researchers follow Wei to data and testing suppliers, and whether Mercor can turn star hires into enterprise contracts. For Meta, the test is whether it can keep the team it paid for long enough to ship.