When OpenAI Chief Technology Officer Mira Murati left to launch a startup last year, it was a page straight out of Silicon Valley’s acclaimed entrepreneurial playbook. A few months later, when 19 current and former OpenAI employees — including co-founder John Schulman — announced they had joined Murati’s Thinking Machines, it was a vivid display of another Silicon Valley tradition: the heated talent wars that accompany each new wave of technological progress.
As AI-mania sweeps the tech industry and reshapes business strategies, the race is on to lock down the specialists with the coveted technological skills and know-how. Recruitment raids like Murati’s on her former employer epitomize the high-stakes maneuverings that have emerged over the past 24 months as companies track top AI researchers and engineers like prized assets on the battlefield. The most intense battle is over a small pool of AI research scientists — estimated to be fewer than 1,000 individuals worldwide, according to several industry insiders Fortune spoke with — with the qualifications to build today’s most advanced large language models.
Stock grants for these scientists can range between $2 million to $4 million at a Series D startup, says Tim Tully, a partner at venture capital firm Menlo Ventures. “That was unfathomable when I was hiring research scientists four years ago,” said Tully, whose October AI talent survey found that AI research scientists working on foundational AI research and theoretical advancements hold golden tickets into top-tier companies.
CEOs like Meta’s Mark Zuckerberg are rolling up their sleeves to personally woo AI stars, according to reports, and influential VC investors like Thrive Capital’s Joshua Kushner have had to run defense, seeking to persuade OpenAI employees of the economic advantage of staying put. (Kushner did not return requests for comment.) To compete with larger, public companies, startups are regularly holding liquidity events with private buyers that give employees a quicker way to cash out some of the equity showered on them.
Beyond the financial incentives, personal ties to leading figures and adherence to differing philosophies about artificial intelligence have lent a tribal element to Silicon Valley’s latest talent wars. From open source to AI safety and security, the values espoused by the various players, as well as the reputation enjoyed by their leaders, have rewritten the recruiting ground rules and spawned a slew of richly-funded startups led by venerated founders like Murati, her former OpenAI colleague Ilya Sutskever, and Stanford’s Fei-Fei Li, known as the “godmother of AI.”
The imbalance between the scarce supply of AI talent and the acute demand for it at Big Tech companies and startups alike, poses a looming predicament for the fast-growing AI industry. As tech companies and investors pump tens of billions of dollars into chips and data centers to power AI services, and hunt for data and energy to train ever more powerful AI models, the ability to attract a relatively small cohort of people is adding an unpredictable but potentially crucial wrinkle to the competition.
More liquidity, fewer cafeterias
The last time Silicon Valley went to war over engineering talent in the 2010s, Google and Facebook waged an all-out perks arms race, each company seeking to outdo the other with on-campus sushi bars, baristas, and other free restaurants, on-site massage therapists, and fitness and martial arts classes.
Fifteen years and thousands of layoffs later, the fight for talent is no longer being waged with cafeteria menus. Today’s coveted AI hotshots are a smaller, more specialized bunch, and companies are competing for them in other ways.
While the company mission and team of colleagues have always been important draws in tech recruiting, in the field of AI, where practitioners love to debate everything from artificial general intelligence (AGI) to open source models, the right fit is critical.
Anthropic, one of the first AI startups to peel off from OpenAI, was born because of the founders’ disagreement with OpenAI’s approach to safety and commercialization of the technology. Many of those who left OpenAI for Murati’s Thinking Machines were drawn by their loyalty to the Albania-born founder (and perhaps to the stock options that can reportedly be sold with nearly no waiting period and at a near zero strike price).
Cristobal Valenzuela, co-founder and CEO of AI video startup Runway, stressed the importance of the right cultural and mission fit. “Our specialty, and the people we have attracted who have come from different research labs to Runway, are people who care a lot about the intersection of art and science,” he said, adding that in a few cases candidates he had lost to other companies returned to Runway because they realized it was the right choice.
“The most talented people want to work on big problems and move the needle,” said Rob Biederman, managing partner at VC firm Asymmetric. That can often be an advantage to startups and play against the established tech giants, he said: “Nothing against Apple specifically, but I wonder if in the context of that organization, do you have the opportunity to move the needle?”
Apple did not respond to a request to answer that question. The company appears to be having some success luring AI talent – at least when it comes to poaching from Big Tech rival Google. Since 2018, Apple has attracted at least 36 AI specialists from Google, including, for a time, Ian Goodfellow – who then returned to Google in 2022 as a research scientist at DeepMind. For its part, Google has fought hard to bolster its ranks amid the competition. Noam Shazeer, one of the co-authors of the AI industry’s seminal Transformers paper, who left his role as a software engineer at Google in 2021 to found Character AI, was hired back to Google, along with several members of his team, in a 2024 “acqui-hire” deal to the tune of $2 billion.
As more AI startups jump into the fray, the supply and demand imbalance is getting worse. In addition to the big AI model companies like OpenAI, Google, Meta, Anthropic, and Amazon, there are dozens of well-funded late-stage AI startups, including Databricks, Perplexity, Glean, Harvey, and Writer, as well as hot newcomers like Elon Musk’s X.AI and Sutskever’s Safe Superintelligence.
To compete with the big fish, privately held startups are offering employees opportunities to cash in some of the paper wealth tied up in their equity. Umesh Padval, a managing director at Thomvest Ventures, said OpenAI has been particularly active on this front, with multiple tender offers, which give employees the ability to sell some of their shares to private buyers. Typically, employees at startups or private companies have had to wait for an IPO or acquisition to cash out their equity.
“This is unique,” said Padval, adding that now other AI companies have begun adopting the same tactic to stay competitive in hiring by making their equity offers more attractive.
Cash compensation is also soaring, even for junior and mid-level AI talent. “We’ve seen a lot of startup offers in AGI and robotics where it used to be $250K for your base, but it’s more competitive now,” said Garett Gentry, a recruiter of AI talent who has worked in-house at Meta, Palantir, Google and Amazon. “I’m seeing offers of $350,000 for a base or more, which would be roughly equivalent to a PhD plus five years coming from one of the FAANG labs.”
The rise of the academics
The AI research scientists that are now most in demand have historically come from academia—mainly PhDs in computer science, mathematics, statistics, or neuroscience. They were typically not the highest paid within tech companies, said Menlo Ventures’ Tully — the big bucks went to software engineers who shipped product.
Today, the tables have turned, and research scientists are the stars of the show. Their path from academia to the tech industry was laid as tech companies funded AI research at universities and more recently have offered hefty compensation packages to bring PhDs — and their teams — in house.
“You can think of senior academics from UC Berkeley who’ve gone into Bay Area tech companies like Databricks and brought a lot of their students,” said Gentry. When Stanford professor Fei-Fei Li recently co-founded AI startup World Labs with her former student Justin Johnson, he suggested two other founding team members: Christoph Lassner, who had worked at Amazon and Meta’s Reality Labs; and Ben Mildenhall, who had created a powerful technique called NeRF as a senior research scientist at Google.
“Besides the PhD advisor type of mentorship or intellectual leadership relationship, AI/scientific research is inherently social in many ways,” Gentry said in an email, describing the tendency for researchers of a feather to flock together. “While individual scientists undertake experiments, the broader context of research is fundamentally collaborative: peer-reviewed journals, sharing/testing ideas with the broader scientific community, the desire for societal impact, the broader ethical context.”
That ethos can present a challenge for tech companies with a penchant for secrecy. While many AI companies, including OpenAI and Anthropic, allow employees to publish research and present at conferences on topics like AI safety and security, there are typically restrictions around publishing research that relates to the inner workings of the AI models due to concerns about competition and intellectual property. Meta has tried to turn this to its advantage. In an interview with Fortune last year, Yann LeCun, chief scientist at Meta and founder of Meta’s Fundamental AI Research (FAIR) lab, cited the fact that the group publishes its research openly and extensively as “a huge argument” for scientists to join Meta.
“If you tell scientists ‘Come work for us but you can’t talk about what you do’…they sense that they will be in a kind of golden prison [where] they will lose their reputation for research,” LeCun said.
A good recent example of this is Nicholas Carlini, a Google DeepMind researcher who announced last week that he was joining Anthropic – but with only a year-long commitment. He had joined Google Brain after earning his PhD in 2018, which he said in a career update blog post last week was his “dream job.” But he said that the process of getting papers approved for publication in journals over the past several years “has gotten considerably more difficult than when I first joined.” The only way he has been able to publish the papers he has, he added, is “because of my willingness to just force my way through, regardless of what the rules say.”
Recruiting – and retaining AI talent – is key
Runway CEO Valenzuela said he spends most of his time recruiting. “I still interview everyone, I still reach out to the best people that I want to work with,” he said, adding that it sometimes takes a couple of years to hire the talent he wants most. “I just had an interview in the morning with someone who we’ve been studying for a year and a half, from a large research lab,” he said. “We’re trying to close that deal. I hope.”
VC firms are heavily involved in recruiting, with talent teams doing networking and interviews. Biederman said he works hard at building a constant talent pipeline. “As a VC your job is basically being a talent scout,” he said.
Gantry, who specializes in hiring scientists and engineers with expertise in machine learning, computer vision, robotics, and other AI domains, said his scouting takes him around the world, including top academic research conferences. “There’s kind of the standard calendar,” he said. “It’s important to go to those to keep up on the state of the art, to participate in the community, to see what’s new – those are places where you meet people.”
Some AI startups are tackling the talent wars by going where elite talent lives – and offering them the unique opportunity to work closely with their compatriots. Ilya Sutskever’s Safe Superintelligence, for example, reportedly has only 20 employees (none of which share their employment status on social media) including a half-dozen researchers in a recently-opened office in Tel Aviv. According to reports, the Israeli employee search has been conducted by word of mouth between friends and former army colleagues. Several of the recruits come from Google Research’s Tel Aviv outpost or top Israeli universities – and were recruited for their math and physics expertise.
Moshe Shalev, co-founder and chief product officer at Tel Aviv-based AI model startup Decart, which is already valued at $500 million and focuses on speeding up AI training, said that there is a fierce battle for the best local talent – with companies all working to recruit from the same pool of AI researchers. The country is working to feed the pipeline, he explained, by drawing not only from universities and tech companies but from math and computer science experts who have served in Unit 8200, the Israel Defense Force’s cybersecurity unit. “We got into this game a little bit late, but it’s not too late,” he said, adding that Decart has recruited more than 18 AI researchers in Israel, including former Google and Apple employees.
No sign of AI talent battle letting up
At the moment, there are no signs of the battle for elite AI talent letting up anytime soon, or that any one company – startup or tech giant – has an insurmountable advantage. If anything, the competition is only growing fiercer as more startups splinter off from the AI leaders and the pace of AI breakthroughs continues to speed up. That means the brains behind those breakthroughs become even more valuable.
“The more time you spend in tech, the talent is the only thing that matters,” said Asymmetric’s Biederman. “The best companies monopolize talent.”