AI researchers are negotiating multi-million pay packages, just like NBA stars

SAN FRANCISCO – Over the summer, Mr Matt Deitke got a phone call from Mr Mark Zuckerberg, Meta’s chief executive.

Mr Zuckerberg wanted Mr Deitke, a 24-year-old artificial intelligence (AI) researcher who had recently helped found a start-up, to join Meta’s research effort dedicated to “superintelligence” – a technology that could hypothetically exceed the human brain.

The company promised him around US$125 million (S$162 million) in stock and cash over four years if he came aboard. The offer was not enough to lure Mr Deitke, who wanted to stick with his start-up, two people with knowledge of the talks said. He turned Mr Zuckerberg down.

So Mr Zuckerberg personally met Mr Deitke. Then Meta returned with a revised offer of around US$250 million over four years, with potentially up to US$100 million of that to be paid in the first year, the people said.

The compensation jump was so startling that Mr Deitke asked his peers what to do. After many discussions, some of them urged him to take the deal – which he did.

Silicon Valley’s AI talent wars have become so frenzied – and so outlandish – that they increasingly resemble the stratospheric market for National Basketball Association (NBA) stars.

Young AI researchers are being recruited as if they are Steph Curry or LeBron James, with nine-figure compensation deals structured to be paid out over several years. To navigate the froth, many of the 20-somethings have turned to unofficial agents and entourages to strategise. They are playing hardball with the companies to get top dollar, much as basketball players shop for the best deals from teams.

The difference is that, unlike NBA teams, deep-pocketed AI companies like Meta, OpenAI and Google have no salary caps, making the battles for AI talent even wilder. American professional basketballer Curry’s most recent four-year contract with the Golden State Warriors was US$35 million less than Mr Deitke’s deal with Meta.

Over the past few weeks, recruiting AI free agents has become a spectacle on social media, much like the period before a trade deadline in sports.

As Meta, Microsoft, Google and OpenAI have poached employees from one another, job announcements have been posted online with graphics resembling major sports trades, made by online streaming outlet TBPN, which hosts an ESPN-like show about the tech and business world.

“BREAKING: Microsoft has poached over 20 staff members from DeepMind over the last six months,” reads one recent TBPN post about Microsoft’s hiring from Google’s DeepMind lab.

Mr Jordi Hays, a co-host of TBPN, said that as tech and AI have gone mainstream, more people are following the recruitment fray “the way our friends from college obsess over sports – the personalities, the players, the leagues”.

On July 30, Mr Zuckerberg said Meta planned to continue throwing money at AI talent “because we have conviction that superintelligence is going to improve every aspect of what we do”.

Superintelligent AI would not just improve the company’s business, he said, but would also become a personal tool that “has the potential to begin an exciting new era of individual empowerment”.

A Meta spokesperson declined to comment. Mr Deitke did not respond to a request for comment.

The job market for AI researchers has long had parallels to professional sports. In 2012, after three academics at the University at Toronto published a research paper describing a seminal AI system that could recognise objects like flowers and cars, they auctioned themselves off to the highest corporate bidder – Google – for US$44 million. That kicked off a race for talent across the tech industry.

By 2014, Dr Peter Lee, Microsoft’s head of research, was likening the market to that for up-and-coming pro football players, many of whom were making about US$1 million a year.

“Last year, the cost of a top, world-class deep learning expert was about the same as a top NFL quarterback prospect,” Dr Lee told Bloomberg Businessweek at the time, referring to a type of AI specialist. “The cost of that talent is pretty remarkable.”

The leverage that AI researchers have in negotiating job terms has only increased since OpenAI released its ChatGPT chatbot in 2022, setting off a race to lead the technology.

They have been aided by scarcity: Only a small pool of people have the technical know-how and experience to work on advanced AI systems.

That is because AI is built differently from traditional software. These systems learn by analysing enormous amounts of digital data. Few researchers have experience with the most advanced systems, which require giant pools of computing power available to only a handful of companies.

The result has been a fresh talent war, with compensation soaring into the hundreds of millions of dollars a year, from millions of dollars a year.

In April, Mr Zuckerberg – whose company was struggling to advance its AI research – dived in by sending personal messages to potential recruits, offering them larger and larger sums.

His approach was similar to that of sports franchise owners, two Meta employees said. Even if the offers seemed absurd, if the new hires could help increase revenue by even half a percentage point – especially for a company that is closing in on a US$2 trillion market capitalisation – it would be worth it, the people said.

“If I’m Zuck, and I’m spending US$80 billion in one year on capital expenditures alone, is it worth kicking in another US$5 billion or more to acquire a truly world-class team to bring the company to the next level?” TBPN’s Mr Hays said. “The answer is obviously yes.”

Meta’s initial offers to engineers varied but hovered in the mid-tens of millions of dollars, three people familiar with the process said.

The company also offered recruits something that was arguably more attractive than money: computing power. Some potential hires were told they would be allotted 30,000 graphical processing units, or GPUs, for their AI research, one of the people said. GPUs, which are powerful chips ideal for running the calculations that fuel AI, are highly coveted.

Mr Zuckerberg has hired with the help of the List, a document with the names of the top minds in AI, two people familiar with the effort said. Many on the List have three main qualifications: a doctorate in an AI-related field, experience at a top lab and contributions to AI research breakthroughs, one of the people said.

The Wall Street Journal previously reported some details of the List.

Some researchers on the List have created chat groups on communications platforms Slack and Discord to discuss offers, two people in the groups said. When someone lands an offer, they can drop the details in the group chats and ask peers to weigh in. AI is a tight-knit field where people often know one another. They trade information about which companies to go to for another offer so they can build up their price, the people said.

Working with friends can be just as important as the money. After a researcher joins a new lab, the first thing that person often does is try to recruit friends, two people familiar with the process said.

The talent wars have started causing pain. OpenAI has changed its compensation structure to account for the shift in the market, employees at the company said, and is asking those approached by competitors to consult executives before immediately accepting offers.

“Are we countering? Yes,” said Mr Mark Chen, OpenAI’s chief research officer, at a company meeting in July, according to a recording reviewed by The New York Times.

But he added that OpenAI had not matched Meta’s offers because “I personally think that in order to work here, you have to believe in the upside of OpenAI”.

OpenAI declined to comment. (The Times has sued OpenAI and Microsoft, claiming copyright infringement in relation to news content related to AI systems. The two companies have denied the claims.)

Not all of Meta’s overtures have succeeded. The company has been rebuffed by some researchers, two people said, partly because Mr Zuckerberg’s vision for AI was unclear compared with those at other companies.

Still, the frenzy has allowed even little-known researchers like Mr Deitke to chart their own destinies.

Mr Deitke, who recently dropped out of a computer science PhD programme at the University of Washington, had moonlighted at a Seattle AI lab called the Allen Institute for Artificial Intelligence.

There, he led the development of a project called Molmo, an AI chatbot that juggles images, sounds and text – the kind of system that Meta is trying to build.

In November 2024, Mr Deitke and several Allen Institute colleagues founded Vercept, a start-up that is trying to build AI agents, which can use other software on the internet to autonomously perform tasks. With about 10 employees, Vercept has raised US$16.5 million from investors such as former Google chief executive Eric Schmidt.

Then came Mr Deitke’s back-and-forth with Mr Zuckerberg. After Mr Deitke accepted Meta’s roughly US$250 million four-year offer, Vercept’s chief executive posted on social media: “We look forward to joining Matt on his private island next year.” NYTIMES