AI Labs: Zuckerberg’s $100bn gamble

0:01Last summer, Mark Zuckerberg was on the hunt for top AI talent. He decided to transform Meta into an AI company, and he needed the best people to make that happen. Meta was offering massive pay deals to AI researchers, sometimes worth hundreds of millions of dollars. But Zuckerberg also had something else up his sleeve. Let's not forget the homemade soup. My understanding is Mark Zuckerberg taking homemade soup to a person who was unwell. We had reported on that and I will exclusively reveal that I'm 99.9 % certain the soup was homemade by Mark himself. That's CFT's Hannah Murphy. According to her reporting, a researcher at OpenAI was weighing a bumper offer from Meta when he fell ill.

0:52Mark Zuckerberg himself turned up on his doorstep, chicken soup in hand.

0:57Cristina Criddle: I have a source on this as well. Christina Criddle also covers AI for the FT in San Francisco. The story I've heard from the OpenAI side is the head of research at OpenAI, Mark Chen, said that Meta had approached nearly half of the people who were reporting directly to them and that there was this fund for around$10 billion specifically for talent acquisition. and he'd heard this soup story that it was handmade soup delivered by Mark Zuckerberg himself. Clearly, when you have that much money on the table, a little bit of food can help to sweeten the deal too. Mark Zuckerberg built the world's biggest social media company.

1:36And on top of it, an advertising empire worth hundreds of billions of dollars a year. Then he performed one of the biggest pivots in corporate history, deciding that the future of technology lay in the metaverse. But now he's pivoting again, committing tens of billions to turning Meta into an AI company. It's a huge transformation and a huge gamble. So will Zuckerberg's money and his homemade chicken soup be enough to pull it off?

2:11This is Tectonic from the Financial Times. I'm Murad Ahmed, the FT's technology news editor. A handful of Silicon Valley companies are vying to lead the world in AI. In this season, I've been talking to the FT's expert reporters about each of them. So far, we've assessed the chances of OpenAI, Anthropic, XAI and Google DeepMind. In this final episode in the season, can Mark Zuckerberg's billions win him a place at AI's top table?

2:48To talk about Mark Zuckerberg's AI ambitions at Meta, I spoke to Christina Criddle and Hannah Murphy. I started by asking Hannah about Meta's latest pivot. Given how serious Zuckerberg seemed about the metaverse just a few years ago, how serious was he now about turning Meta into an AI company? I mean, it was already fairly dramatic turn of events when in 2021, I think it was, Mark renamed the company Meta, announced this long-term vision of becoming a metaverse company and tens of billions of dollars being invested into this idea that we will have an avatar-filled virtual space where we'll all hang out and interact.

3:33That is nothing compared to how dramatic this latest push is from Mark. We're talking hundreds of billions of dollars into becoming an AI company, reshifting and reshaping the entire company structure around AI, bringing in new leaders, rounds of layoffs. I mean, this is very, very, very, very dramatic. And he's sort of betting the farm on this new vision several years after the metaverse vision, perhaps didn't pan out as planned. And how can they afford this? Well, the ads business, a very well-oiled machine that Mark built with Sheryl Sandberg, remains strong and in fact is forecast to take over Google by the end of the year by eMarketer.

4:23So that is there. He has this cash that he's sitting on and so he can afford to do it. At least he's made a calculated bet. And in interviews, he said the cost of not investing in AI is more dear than the cost of throwing money at it. And perhaps I lose some billions here and there. That's sort of how he's framing it. Christina, what do you think? You think I've

4:46Cristina Criddle: captured that? I was just going to say, let's not forget that Meta has actually been in AI for a very long time. Like it had its fundamental AI research lab set up since 2013. And the thing that's driving this big cash machine is all AI. The whole reason we have this feed on our Instagram pages that's personalized to us that keeps us engaged is because of AI. And it's because an AI is determining what we're going to be interested in, what we're going to engage in. It's just not the sexy kind of AI that we have now. And so I think when Mark saw the ChatGPT revolution happening, he saw other companies kind of taking a bit of space and more ownership of AI, he was upset.

5:31Cristina Criddle: And a highly competitive man will just push everything he has towards winning that race. And so I think that's what you're seeing now is they were always in AI, but now more than ever, Marx realized, okay, well, we need to be at the forefront of this type of AI, this generative AI revolution. But as you say, that massive ads business gives Meta a lot of money. And one area where they're spending a lot of money is trying to recruit AI talent. There's a relatively small number of AI researchers that you need to do the cutting edge AI work. How's that been playing out in the valley? Yeah, I think we've spoken in previous episodes about the kind of pull of the different labs, like Anthropic, which have this ideological motivation to create AI that is safe and beneficial for the world.

6:20Cristina Criddle: Meta doesn't necessarily have that. And it does kind of have a bad rep from the social media era. But what it does have is lots of cash. And so I'm hearing of these crazy offers being put on the table. I have people I've spoken to who have had offers from Meta, and they've basically said, look, like we have families, we have to make a living, we're already paid quite well. But when you see millions on the table, how are you going to go home and tell your partner that you've turned that down? And the other really fun detail on this is how much Zuckerberg has been a part of it. He's been personally approaching, texting people, hosting researchers at his home, and even sending them soup to try and win them over.

7:06Hannah, I think that what all of this gets to is the sense that Meta feel like they have to buy their way in here. Their inherent pitch is weaker than everybody else's, or at least they feel like they need to do things that are above and beyond to ensure that they get the possible talent. One name they have been able to attract is Alex Wang, this billionaire AI startup founder still in his 20s. Who is he and why is he so important to Meta? Alex Wang is the man who Mark has brought in to head up most of his AI efforts. He heads up a team called Meta Super Intelligence Labs, or MSL, which is most of the efforts to build AI products and AI models.

7:50Within that, there's a group of about 50 or so engineers who are called the TBD Lab, To Be Determined Lab is the acronym there. Those people are focused on driving and building and developing the frontier models, which will then be embedded into the products. So they're really at the bleeding edge. They're the very highly sought after AI researchers. Alex previously built a startup called Scale AI that's focused on data labeling for AI labs. And he therefore had all these relationships with the various labs, very, very well connected, known as something of a socialite around Silicon Valley, and so has been able to help bring in that talent, but also has this frenetic, ever sort of testing and trying energy that has helped propel Meta forward and that certainly Mark hopes will bring Meta into the current AI age.

8:48Christina, you had covered scale as a startup. What was his reputation throughout AI circles? Is this the guy that many others would have picked out to be the person that you absolutely have to have to drag your big tech company into the AI future?

9:05Cristina Criddle: Well, I think what you've seen is a complete culture switch in who's running AI at Meta. I mean, you've seen the loss of Jan LeCun, who's one of the godfathers of AI, Joelle Pinault, who's very respected Canadian AI scientist. These people have clout. They've been in the industry for decades and they're out. Former Meta executives and now they've left. They've left. And there's this new guard of very young AI researchers which want to win. And they're very motivated to work hard, maybe do less of the kind of foundational, scientific, grounding AI research that's exploratory. And actually work on shipping a model every few weeks to keep pace with the other AI companies.

9:54Okay, so we've talked about who they've hired, how disruptive it's been, the culture shift within the company, but we haven't really discussed the most important thing of all. What on earth are they building? What is Mark Zuckerberg's AI vision here? Let's get to that after the break.

10:23retirement planning. And to think, that's just a small taste of what Schwab offers. Because Schwab knows that when it comes to your finances, choice matters. No matter your goals, investing style, life stage, or experience, Schwab has everything you need all in one place. So you can invest your way. Visit schwab.com to learn more.

10:47Hannah, we've talked about how Meta has performed this massive pivot to AI, but what is it trying to build? What is Mark Zuckerberg's vision when it comes to AI? We are still beginning to work that out. About a year ago, Meta put out a memo that said, we want to build personal superintelligence. And since then, it's slowly begun to emerge what exactly he means by that. Personal superintelligence, he said, you know what, we want to focus on what can be most useful to you and free up your time, open up your creativity. This is the sort of language he was using a year ago. Over time, we've seen that through some of the products, early products before this new cracked team of engineers came in.

11:33Meta introduced its own chatbot, Meta AI, an assistant, and it also introduced some characters. So you could chat with a sports expert type character or a chef type character. Some of these were based on celebrities. They also created a way that you could create your own user-generated characters. So there, there's a sort of companionship angle. And Mark said publicly, I think, you know, most people have a handful of friends. Actually, you want more like 14 friends and AI can help you get there. As many as 14 friends. As many as 14 close friends. However, since then, and since Alex's team has been set up, and also there's been sort of a movement on the technological front around AI, the recent rhetoric coming from Mark and Alex has been around agents.

12:21So both having personal agents, AI agents, and having business AI agents. They're looking into an agentic AI assistant that could go carry out tasks for you, do shopping for you, go book a flight for you, for example. So this world of AI agents is coming from reading between the tea leaves of what they've been saying in various interviews and earnings calls. And then further down the line, two areas that Meta is exploring. One is whether there is a new form factor to use their lingo that we might use to interact with AI, be that glasses. So they already have these meta Ray-Ban glasses that have cameras, that have speakers, and that now they're looking at a little display within the Ray-Bans that you'd be able to sort of verbally ask questions of AI and in the display, it might answer those questions.

13:14So that sort of journey towards augmented reality glasses, the very final area being robotics. If we are going down the route of agentic AI, is there a way in which we can have a physical robot then be able to carry out some of those tasks as well? Christina, what do you make of this personal superintelligence concept? I was in a recent meeting with a big senior figure within a major AI lab, and they really scoffed at Meta's efforts here. Everybody else was shown a degree of respect, but it was quite noticeable that they saw this concept of personal superintelligence as something not to be taken too seriously.

13:59I don't know how seriously we should take it.

14:01Cristina Criddle: It's just another buzzword to me. I think we've heard a lot in the AI field about AGI, artificial general intelligence, or this sense that one day we might have AI systems that are as intelligent or more intelligent than us. And lots of the labs, lots of the big tech companies are all working towards this goal. Some, I think, more credibly than others. I think what's interesting is that Meta in its AI play has mentioned all of the same buzzwords as all of the other competitors. So AGI or super intelligence, whatever you want to call it, coding models, frontier models, agents, all this stuff. But actually in amongst that, I can really see some clear business lines here for Meta.

14:42Cristina Criddle: What it does have is distribution through Instagram, Facebook, WhatsApp. It knows consumers well. And so you can see that in the companionship offering it has, that it might be trying to play off people's need to engage and engage on these platforms. And then also you can see that with the glasses that Hannah mentioned, this sense that they have all this audiovisual data already. And one of the things that Meta's model is actually good at is audiovisual. And so they have all this proprietary data and all this user behavior that I'm sure OpenAI or Anthropic would love to have. So it makes sense to me that they would try and push on that and try and use the data that they have, the user base that they have, and try and put AI in front of those people in a way that generates returns.

15:33Cristina Criddle: And actually, that could be great news for Meta. It does have that advantage. OpenAI and Anthropic are a bit more of a blank slate. And those returns could come from a subscription model, right, Christina? Or that it feeds back into the ads model in some way that if we're sending off our agent to go do something for us, it will show that agent and ad at a certain point. They'll get more data about sort of shopping intent, for example, and be able just to place regular ads across the platforms. For me, the question is how they manage to integrate their offerings into the surfaces they have in such a way that they're sticky for users.

16:13I think To be critical, it hasn't worked when it's come to WhatsApp. For example, they put a Meta AI box at the top that is the same box as you might type to look within your messages for an old message. And you go to look for something and then you've accidentally generated an AI query that's annoyed users.

16:33Cristina Criddle: And they use that for advertising as well. So that's what's really annoyed me is every time I go and search for something in my chats, this is as a user rather than as a reporter. I'm concerned that whatever I'm typing into my search bar is going to be used to serve me advertising. Right. So I think some of the big questions are how they manage to integrate these products into their existing product ecosystem. And then also they have to be conscious of the trust, the privacy part, where exactly is this data going? They've had a reputation in the past for being somewhat lax with privacy. So they're going to have to make sure that whatever they do, people feel comfortable, especially at a time when young people in particular are very nervous about AI and almost to the point that they're sort of completely skeptical and trying to reject it.

17:21At the same time, as I said, Meta has these very compelling, addictive social network with a lot of young people using. So there is an opportunity there, but they have a very delicate line to talk. This has been a long-running problem with Meta and how it's developed over time, right? They've got all these advantages. They've got huge amounts of money to build the computing capacity needed to build models at the frontier. They have some of the best data sources possible that's been developed over many, many years. But they have this creepy problem where people feel that their data is being used in ways that they don't want to particularly.

18:02they blurred the boundaries between what's personal and what is commercial you can see even in the glasses that they've made with ray-ban they're very popular they've actually managed to find a wearable product that is doing very well but it's led to a lot of concerns that a lot of people are using this to essentially spy on other people take videos without consent all this other stuff and they kind of haven't shown that they care enough about these issues how do they get over the problem of being considered to be the creepy company that they really care about these issues of privacy?

18:38Cristina Criddle: I might push back on this a little bit, actually, because I think, as you said, like the Ray-Ban Metas have been really popular. I keep seeing people wearing them. And I think we do live in this age, whether you like it or not, where people are filming and capturing content all the time without our permission in public. And Meta played a big role in creating that world, but so did the iPhone cameras generally and TikTok, all of the social media that we have. And so I think our sense of creepy, yes, in my opinion, it's creepy, but it's kind of irrelevant. People are going to be wearing these things and doing that anyway.

19:15Cristina Criddle: I'm not seeing a huge effort to try and regulate that around the world. And I do think there is this demand to have this kind of hands-free access to AI and digital content where we're not attached to our phones. And so I do see the glasses as being quite forward and a potential win for meta. You've seen Google try and get in this game as well. If we can crack a form factor that might be glasses, where we can speak to our AI assistants, where we can be less attached to our phones and maybe engage in the world a bit more. I don't necessarily know that that's a bad thing, but will it mean that perhaps we're all being watched a bit more?

19:57Cristina Criddle: Probably. I mean, this makes me feel miserable. If I'm honest, the impression I'm getting of Meta is a company that's willing to use its advantages to get to the front of the race, but kind of do it in a way that doesn't really take into account what's good for us, what's good for the world? I think it's hard to know at this stage. I mean, a personal assistant that helps us do everything, if it's safe, their narrative is that it will free up more time for us to be more sociable and enjoy our lives and be more productive. And to some extent, that might be correct. The counterpoint is also that actually, you know, they started off as a social media company.

20:38You could argue that it's not social to be interacting with a bot all day. And that if those bots get very addictive on the companionship side, that's sort of antisocial rather than social and isolating. Again, on the agentic side, if it's doing tasks for you, you then are free to do more out in the world and be more sociable. So there's a tension there and it depends how people end up using these. And without the models being as advanced as they will be, it's hard to assess exactly how they will end up being used and to what extent we do end up relying on them for tasks or for socializing.

21:18Cristina Criddle: And I think the real question here is a power one. Like I remember when I was covering Meta or Facebook at the time, before all of this generative AI, I spoke to somebody who was on a team who was dedicated to deciding, okay, what is actually good with AI? How does AI determine what is good? And who are we to determine what is good in the world as well? And that question has stayed with me throughout. Who are Meta or Mark Zuckerberg? Who are they to say what's good and what is a good way to use AI for society? Who is open AI or anthropic to do that as well? And I think it all comes back to power.

21:57Cristina Criddle: These companies have the power to bring this technology to the world and to everyone, but should they have that power? And do we trust that they will do the good thing? And I think one other area that Meta has to navigate beyond the sort of privacy trust piece is that being a group that's placing such a huge bet on the infrastructure piece, we're talking hundreds of billions of dollars, data centers the size of Manhattan, buying up land, doing energy deals of all kinds, really, really throwing the book of this. There's now beginning to be more backlash against data centers, what they do to the local economy, local environment, from a climate change perspective, a lot of nimbyism, not in my backyard.

22:45So that is a growing debate that I think is only going to get louder. So we talked about this concept of personal superintelligence, but it seems like a lot of this is personal to Mark Zuckerberg himself. Hannah, you've done many years of reporting on Meta now, and you've been particularly strong on the kind of the hopes and dreams of Mark Zuckerberg himself. Why does he want to be in this race? And why do you feel like he wants to win quite so badly? I think Mark personally wants to be seen as a visionary. He wants to be in among the recognized leaders. He's very competitive. My understanding is, you know, Elon Musk came in, took over Twitter, gutted the company, didn't care what people thought.

23:32And there's some of that energy that Mark, that sort of more unapologetic energy that Mark is bringing now to Meta after years of perhaps, you know, he really got an onslaught of criticism from the kind of 2016 through 2022 period. And Metta made some efforts in earnest to address some of that criticism. And still, when the Biden administration came in last, still got rejected. He did not get a seat at the table with other AI companies with the Biden administration, was not taken seriously as an AI player. I think he wants to put Metta back on the map. And he sees this as a new opportunity, I'm sure, for a reset in reputation, but also sort of technologically speaking, a reset in reputation after the metaverse perhaps didn't gain quite the traction expected at quite the rate hoped for.

24:27I'd add one more thing, which is that there is an argument that Mark perhaps overcorrects, that when he feels he's behind or when he does something, you know, he does bet the house on it, but he perhaps goes too far one way. And there's these massive swings, swing towards the metaverse, swing towards AI. Some people will see that as bold and managing to keep the company in the 21st century. Others say, you know, that creates unnecessary issues, lose money, share price up and down. So it's yet to be seen whether this is quite the overcorrection, especially with some of those infrastructure bets.

25:03It's hard to tell without a crystal ball, but it's certainly big, distinctive bets that he's making. So, Christina, are you a bull or a bear when it comes to Meta? Do you think that they've got a good shot at being the winner of this race or, you know, having a podium finish?

25:21Cristina Criddle: I don't think Meta is going to be the winner. I'm not even sure that there is going to be like one winner. I don't think it's going to completely evaporate into nothing. Meta has the cash. It has the compute and it has this leader who is like a king in control of his kingdom and worried about losing it. So it's going to try really hard to make this AI play work. And it does have these very powerful social media platforms, a huge cash machine with advertising and products like the glasses, which are showing potential. So I don't think we can discount it. However, on the frontier model AI front, it's quite late to this.

26:02Cristina Criddle: and it hasn't managed to catch up so far. Its open source strategy seems to have completely disappeared. And so whether it's going to be a front-end company that's creating so-called super intelligence, I doubt that. Hannah? I think it all comes down to how it integrates the models into its products and what exactly the products are and how it monetizes them rather than whether it's at the cutting edge. I wonder whether obviously you need a good, compelling model and they need to keep up to some extent, but do they have to be the very, very cutting edge? Will most consumers know the difference between one that's here and one that's there?

26:42I feel like if they are at least, you know, semi-competitive and it's compelling enough to satisfy most consumers, the question is really how they integrate them and make a consumer experience, right? They're going after a consumer market, not an enterprise market. And how do they make that resonate and make that compelling and make that addictive, make us come back for more and potentially choose, okay, I want to use this AI assistant for ease. It's seamless. It's easy. It does what I tell it to. It knows stuff about me versus a Google one that's integrated versus I have to go log into my OpenAI.

27:19That is, for me, where the edge will be. So we shall see. Christina, Hannah, thank you very much. Thank you. Thanks. That was the FT's Christina Criddle and Hannah Murphy speaking to me about Mark Zuckerberg and Meta's AI ambitions. You can read much more of their reporting on ft.com and have included some free-to-read articles in the show notes. This is the last in the current season of Tectonic. If you haven't listened already, there are four more episodes on the Silicon Valley labs battling each other for AI supremacy. OpenAI, Anthropic, XAI and Google DeepMind. Tectonic will be back in a few weeks, but in the meantime, we want to hear from you.

28:04We're running a short survey of Tectonic listeners. It's your chance to tell us what you think of the show, what you like or don't like, and what you want to hear more of. There's a link in the show notes, and completing the survey gives you the chance to win a pair of high-end Bose headphones. See the show notes for T's and C's.

28:29This season of Tectonic was hosted by me, Murad Ahmed. The producer was Edwin Lane. Topher Forges is an executive producer. Sound design by Breen Turner and Sam Giavinko. Original music by Metaphor Music.

29:02and audit defensively. Avalara, agentic AI tax and compliance with confidence. Capital raising is being redefined in real time from macro disruptions and shifting investor expectations to AI-driven demand and the rise of private credit. When headlines can tilt markets overnight, how can companies stay on course? Strategic Alternatives, the RBC Capital Markets podcast, explores how corporates and investors are navigating new pathways to raise capital, create value and drive growth. When conditions change, alternatives matter. Listen to Strategic Alternatives, available wherever you get your podcasts.