weekend pitch 7/20

Featured image by Josie Doan/PitchBook News

Weekend Analysis

OpenAI has lost over 25% of its key research talent in the last 2 years

Meta and Thinking Machines have aggressively courted researchers from OpenAI.

OpenAI is losing its grip on top AI talent as intensifying competition threatens its first-mover advantage.

A PitchBook analysis found that over 25% of its research leads, or more than 50 key staffers, have left for rival companies or started their own in just the last two years. Their ranks include elite engineers behind leading models like GPT-4 who have left for outsized pay packages at Meta and Thinking Machines.

This exodus could undermine OpenAI’s position as the leading AI innovator, as competitors aggressively court the experts needed to build the next generation of large language models. Despite these challenges, OpenAI is fighting back with new staff retention strategies, including upping compensation for key engineers.

OpenAI did not respond to PitchBook’s request for comment.

The leading AI shops are all vying for the small cohort of engineers trained at top machine learning labs who have the skill sets required to build increasingly sophisticated large language models.

Meta has offered eye-watering compensation packages, in some cases as high as $300 million over four years, to at least 10 OpenAI employees, according to a Wired report. Last month, the social media giant brought on four senior researchers involved in GPT-4: Shengjia Zhao, Shuchao Bi, Jiahui Yu and Hongyu Ren.

But OpenAI, last valued at $300 billion and widely considered the AI player to beat, isn’t going down without a fight. Last month, Wired also reported that OpenAI chief research officer Mark Chen told employees that they had been “recalibrating comp” for top talent. “I feel a visceral feeling right now, as if someone has broken into our home and stolen something,” Chen reportedly told staffers.

Early employees of OpenAI whose stock options have fully vested are in an extraordinary position: Their net worth has skyrocketed well into the millions, and they have the option to decamp to the AI lab of their choosing.

These are not your average ML engineers. “There’s a very small handful of pure researchers who are being treated as almost professional athletes in the way they get compensated and traded around,” said John Loser, GP at Floating Point, an emerging VC firm.

Some top researchers have even hired agents to help them score a better deal. “Folks are bringing on advisors to negotiate comp packages,” according to Marco DeMeireles, co-founder at VC firm Ansa Capital.

As companies race to achieve artificial general intelligence, or AGI, they’re courting VCs for massive amounts of capital to spend on compute and to acquire this rarefied talent set.

Thinking Machines, which is less than a year old, has already raised $2 billion in venture funding.

Meta’s Mark Zuckerberg has said that Meta plans to spend between $60 billion and $65 billion this year on capital expenses and turn Llama into the leading AI assistant on the market.

For smaller startups, this talent set is increasingly out of reach.

“For virtually any kind of seed, Series A company, the idea of paying millions of dollars for a top researcher, no matter how valuable you think they truly could be, it’s just not feasible,” said Loser.

Featured image by Josie Doan/PitchBook News

Learn about our editorial standards.