Why AI Researchers Are Getting Paid Like NBA All-Stars  ## Metadata - Author: [[Decoder with Nilay Patel]] - Full Title: Why AI Researchers Are Getting Paid Like NBA All-Stars - Category: #podcasts - URL: https://share.snipd.com/episode/e5902569-13db-4ff3-aee2-7beecd022cb2 ## Document Notes ## Highlights - **Meta's AI Hiring Deficit** - Meta is aggressively hiring top AI talent because it's behind in AI model performance compared to rivals. - Successful model releases directly impact user engagement and company valuation, making AI progress critical. Transcript: Alex Heath I think it's important really to step back and think about why is Meta doing this specifically? And why are they being so aggressive with going out and trying to hire people and spending amounts that surpass what the highest paid CEOs on the S&P 500 make. And I mean, it's obvious if you follow this, that Meta is behind in AI. Llama 3 was a good model, was well-received. And then Llama 4 earlier this year was not. And actually internally, what I heard was that Llama 4 in the user's eyes was worse than Llama 3. ([Time 0:05:13](https://share.snipd.com/snip/fb14351a-da9a-4d5a-af42-b6cab2b9c93a)) - **Complexity of AI Offers** - Top AI researchers often receive complex offers tied to performance and tenure with clawbacks. - These offers aren't straightforward guaranteed payments, complicating the impression of huge salaries. Transcript: Alex Heath Right. Hayden Field Some of them are tied to specific performance metrics that may be a little bit vague or may be decided upon by the people above you. That's what I'm hearing, at least from people that have turned down these offers. They weren't fully sure on what it would really entail. And so, you know, they would rather just stay where they know what's what. Alex Heath It's important to distinguish the researchers from the engineers, the people who are researching frontier models. There's not a lot of them. This is actually a supply and demand issue. Based on my conversations, it's easy to say that there's probably less than 2000 people in the world helping build a frontier model from the research perspective. So when you have that kind of imbalance of the amount of money that is invested into the space and the people who can actually build it, they become a very, very hot commodity. So that makes these numbers less crazy in the context of the money that's being spent when Meta, Google, Microsoft are spending tens of billions a year on data centers, on data to build These models, spending a few billion on the talent ([Time 0:08:07](https://share.snipd.com/snip/ae0d3afe-b15d-407f-ab4e-af64559a3195)) - **Google's Windsurf Hiring Drama** - Google hired Windsurf's CEO, co-founder, and R&D employees after OpenAI's acquisition deal fell through. - This left Windsurf with a shell company and shows the drama of AI talent poaching. Transcript: Hayden Field We saw OpenAI's deal to buy Windsurf that had been really reported over months and months suddenly off. And instead, Google swooped in and hired the CEO, the co-founder, and a lot of their R&D employees and just brought them onto the Google DeepMind team to build agents. So they're going to be working on Gemini. Google's not going to have a stake in Windsurf or Control, but it's going to take a non-exclusive license to some of Windsurf's tech. So it's been pretty crazy. Their head of business became their interim CEO. They had to suddenly name a new president. And a lot of the employees were kind of shocked by this because they had been pretty close for a long time. It had been a team that had known each other for a while. It was a perfect example of how soap opera-esque these hiring wars are starting to become. You know, when you're at a company and you've been there for a while, maybe you're employee number one, employee number two, you see on Twitter that most of your company, the talent of Your company is being sold to Google and you're just kind of left with a shell. It's really, you know, just, yeah, an example of the time we're in right now. Alex Heath Let's unpack that a little more. So Winsurf is, you know, one of the hottest AI coding startups. They were going to sell to OpenAI, that deal fell apart. And then, like you said, kind of like a soap opera, the team gathers and finds out that Google is actually hiring away the CEO and the core AI engineering team. So 40-ish people, the company is like 250 to 300 people. These people are already leaving. They're already out the door and there's a new CEO and you all figure it out. ([Time 0:15:10](https://share.snipd.com/snip/9d45cd28-89a7-464f-800c-b041d4a5358e)) - **Impact of Reverse Acquihires** - Reverse acquihires by big tech often leave startups struggling or declining. - These rapid, talent-driven deals prioritize speed and can disrupt the startup ecosystem. Transcript: Alex Heath My knowledge, I don't think any of these startups that have been reverse aqua hired so far are doing objectively better than they were before. Big Tech came in and stole the founding team. Character AI is a great example. It was a pretty viral chatbot kind of companion platform created by two early Google senior engineers. And, you know, since those guys went back to Google, you don't really hear about character as much. Scale, which Zuck just hired the CEO and some of the core team from for Meta AI's new lab, just did layoffs. You know, WindSurf had to sell itself like within days from Google sweeping in. So yeah, there's I think, interesting second order effect where big tech is coming in to these often really well-funded, well-positioned startups that are really red hot. They take the team and then the startup is left to pick up the pieces. And in many cases they can't. And again, this is like, this is a brazen, like middle finger to regulators. Scale is a great example. My understanding of the Scale deal and why it happened so fast and was structured that way is Zuck wanted Alexander Wang in the office immediately. And he was in the office within a few days of that deal being announced, working on setting up Meadow's new lab. There's an expediency to how these deals work where the acquirer can get the people in its tent ASAP. ([Time 0:20:16](https://share.snipd.com/snip/e8627eb2-aa5c-4f12-b971-f24a706d783e)) - **Motivations Beyond Money** - Many top AI researchers feel they've 'made enough' money and prioritize mission alignment and work culture. - Personal reputations and ethical alignment often outweigh even massive salary offers. Transcript: Hayden Field Something that I've seen is pretty constant over the last few years in AI reporting, but is more evident now than ever, is that a lot of these key AI engineers or researchers are what you Might call post money, like they make so much that making more isn't in and of itself, a motivator for them. Yes, they're motivated by money. But after a certain point, you know, it's kind of all fake, you know, you're making millions, okay, what's another few million. So they care a lot about how their work is being used, who they're doing it with who they're doing it for, what their company's vision is in terms of accelerationism, you know, societal Impact, safety, risk, a lot of them have differing views on that stuff. But it's important to a lot of them that their company or the leader of the team they work on aligns with them on whatever those views are. So yeah, we're seeing people turn down crazy offers, personal reach outs from execs. You know, I've heard OpenAI CEO Sam Altman always calls at least most technical hires, potential hires with an offer, you know, to personally pitch them. These people know how in demand their work is, you know, because like you said, you know, there's maybe less than 2000 people that can do what they're doing right now and pushing the industry Forward. So they really care about how their stuff's being used, who they're doing it with, what their work-life balance is, maybe not so much the work-life balance, because they're all working A lot, but at least what their work culture is, you know, how happy they are day to day and how much they believe in what they're doing. So that's what I've thought is really interesting. Some of these people won't even consider an offer if they're already pretty aligned with their current workplace and, you know, not much could get them to leave. ([Time 0:27:43](https://share.snipd.com/snip/5c8155b5-8d75-40c9-9532-09b3d5844943)) - **GPU Access as a Hiring Tool** - Access to GPUs is a critical factor for AI researchers when choosing where to work. - Guaranteed compute resources can be as valuable as compensation in attracting top talent. Transcript: Alex Heath A big pitch that these labs make to potential recruits is your access to GPUs and that access being guaranteed. OpenAI has been public about needing to divert GPU allocation away from research to product because ChatGPT has just gone parabolic with its growth. So if you're a leading researcher and what really motivates you is pushing the frontier, achieving AGI, and really pushing forward these benchmarks, you need those GPUs. If someone like Zuck can say, look, I'm building a data center the size of Manhattan, I'm going to guarantee you more GPUs than you can be guaranteed anywhere else. That's a huge draw is what I've heard, because it's basically the GPUs are the fuel for the work. Right. And I don't think that gets talked about a lot. But that to me is just almost as important as how people feel about the mission of the company. Hayden Field Definitely. That's true. Because I mean, people get frustrated when those types of resources are diverted away from their team. There's infighting in that regard. You know, we saw that at OpenAI. ([Time 0:31:00](https://share.snipd.com/snip/5b00aa58-1137-4d0c-bec4-3102801cd7c1)) - **AI Researcher Leverage** - Individual AI researchers hold unprecedented leverage as their ideas can create enormous company value. - This influence is akin to innovations in semiconductors that transformed industries. Transcript: Alex Heath It's a defensive move as much as it's an offensive move. I was talking to an AI exec at Databricks late last year, Naveen Rao, who is always, I think, really prescient on this stuff. And he was telling me something really interesting. I'll just read a quote. He said that the leverage that a researcher has in an organization is unprecedented. One researcher's ideas can completely change the product. That's kind of new. And he compared it to semiconductors, and that being the last area of tech where this kind of leverage from a specific individual was possible, where if you invented a new transistor Architecture, you could potentially create billions and billions of dollars in value. He just, I think, connected the two. And for me, that makes a lot of sense. You know, if you come up with a new scaling paradigm for one of these models that leads to Meta having to spend billions of dollars less on its data centers to run inference for its model, You've paid for yourself, you know, potentially 10x, even if you're making a billion a year. ([Time 0:32:54](https://share.snipd.com/snip/278a4a49-7a2f-4ae4-bd95-8456a5ffcfcd)) - **AI Talent as Superstar Trades** - The AI talent market resembles NBA superstar trades, with researchers receiving huge compensation packages. - These moves garner intense public and media attention, unprecedented for tech hires. Transcript: Hayden Field It is also crazy to see TBPN announcing these researcher moves like their trades or huge athlete movers. I mean, we saw even, and I'm also wondering how these employees are dealing with the limelight. You know, I can't even imagine if I announced on LinkedIn, I was moving to a new media startup, if all these like huge Twitter accounts are like, oh, Hayden Fields moving, like, what's She doing? And, you know, they're maybe not equipped to deal with this, like two Anthropik employees that were not even the top researchers of, you know, building Claude itself, but two leaders Of the coding product, you know, left Anthropik for a big signing bonus, I hear, at AnySphere, the developer of Cursor, but then they came back two weeks later, and it was all over Twitter, You know, every, you know, news account is tweeting about it, like internal tech podcasts. And it's just interesting to see, yeah, these moves reported like they're like, you know, big athlete trades. ([Time 0:36:25](https://share.snipd.com/snip/8eb1253e-20f6-433c-b7d6-d807bcea9e93)) - **Peak AI Talent Bubble?** - The current AI talent bidding frenzy is likely a peak hype, resembling a Hail Mary effort by companies. - Going forward, fewer people will command higher offers due to growing demand for seniority and influence. Transcript: Alex Heath Or do you feel like we're kind of at peak hype here? Hayden Field I think we're at peak hype. Honestly, I think this is like a Hail Mary situation. They're doing all this. They're seeing if it pays off in a few years. And then after that, you know, we see where it goes. I think it will not get bigger. It'll either stay the same or go lower. The other interesting thing to me is that I think as we see this industry progress and advance as quickly as it has been, the kind of amount of seniority you have to have in order to really Be desirable is going to increase and increase. You know, we're already seeing OpenAI and Anthropic talk about how their models are close to or pretty much can already replace an entry-level engineer at their companies. The people that are going to be in demand are those that can shape and guide and deal with the big picture of these models rather than doing entry-level jobs to help train them and, you Know, kind of deal with more of the minutia. So it's going to be interesting to see the titles that are more in demand just kind of keep getting higher and higher rather than, yeah, those kind of entry-level AI hires. I'm glad you brought that up. Alex Heath And that dynamic you just explained is why I kind of agree with you and I kind of don't. I think that there won't be as many people getting these kinds of numbers offers in the AI world, but I think the fewer people who are getting them will get more money because the leverage Continues to grow per individual as AI starts to do more of the underlying work. So maybe it's like 50 people who are each getting $3 billion offers. Honestly, people I've talked to say that is a feasible future to imagine given the rate of change. Hayden Field I can see that. Like you said about individual people being stocks here. Yeah. I mean, you know, they're going to see how that stock performs and then use that to inform future offers. So you're absolutely right. Well, ([Time 0:38:39](https://share.snipd.com/snip/c3f883d9-a058-46d9-b2eb-146bc734fee7))