Chime IPO: Are IPOs Hotter Than Ever?

1. JLJason Lemkin0:00

   There's no way that Scale can recover from losing its founders. I love you, but it's a dead man walking instantly.

2. RORory O’Driscoll0:06

   I could imagine an entire IPO cycle where, once again, neither of the two names you cited chose to go out, and we get to the end of the year.

3. JLJason Lemkin0:12

   I'm starting to lose confidence in the old guard.

4. RORory O’Driscoll0:14

   In situations of ambiguity, the person who has the most leverage has the best chance to win, and I would argue Microsoft has quite a lot of leverage 'cause it's not existential for them, especially now.

5. JLJason Lemkin0:23

   I think they have no leverage, actually, honestly. I'm not trying to be a chaos agent like a prior guest.

6. RORory O’Driscoll0:29

   Ready to go? [upbeat rock music]

7. HSHarry Stebbings0:41

   Guys, I'm so excited for this. It is my favorite time of the week, and oh my God, what a week of news we have to go through. Uh, we're also thrilled to be joined by Garrett from Handshake. So I wanted to start with the news of Scale and the acquisition there. And so I'm gonna start with, obviously, Scale, $14.83 billion to Meta. I wanted to start with that. Garrett, how should we read this? You're the expert in the room. How did you take the news?

8. SPSpeaker1:10

   Well, I think it's in, it's pretty unprecedented, uh, to have the leading player, one of the leading players, like, completely bow out. What we're seeing is, uh, many of the labs having to reallocate their spend and, uh, trying to diversify away from, uh, some of the leading players that, that they no longer can trust to do, to do work right now.

9. JLJason Lemkin1:29

   I have a bunch of questions I want to ask Garrett since we have him. At, at, at, I mean, at a deal level it seems pr- pretty clear. It's a weird deal. They've put almost $15 billion into Scale. Rory may have the details. Most of it comes out as a dividend, and the CEO leaves [laughs] to go run, uh, seemingly a much broader portfolio at Meta, right? He's not just running training, right? Which this is the whole mystery is, um, is that Facebook does, or Meta doesn't even seem to care about the revenue, right? Um, but if only OpenAI is committed to, to maintaining that as a partial presence, I assume the revenue will decline rapidly, right? Um, I don't know, though. That's the... But you've already benefited from that, right? I mean, almost overnight you've benefited from that. So, so in a way, you are, you are the mole here rather than us.

10. RORory O’Driscoll2:13

    Agreed.

11. SPSpeaker2:14

    Yeah, I mean, we have seen, uh, a huge surge in demand. Primary concern right now [laughs] is hiring on our team. I mean, demand tripled in, in the, you know-

12. JLJason Lemkin2:22

    A week

13. SPSpeaker2:23

    ... in, in the aven- of the week.

14. JLJason Lemkin2:25

    [laughs]

15. SPSpeaker2:25

    Uh, I'm running on an average of, like, three and a half hours of sleep for the last, like, 10 days.

16. RORory O’Driscoll2:28

    I think I'm gonna zoom out even one level before, just to restate in case... I mean, we've dived right in 'cause we're so in the details. But zooming out a million miles, yeah, Scale AI is an amazing company founded 10 years ago to do, to help companies build great AI models, and they do that by rounding up human experts and initially just doing very simple data labeling. But increasingly over the last four or five years, as Garrett had said, as the tasks that the model builders have to attempt to solve get more complex, they've been hiring experts and at the PhD level to help with post-training reasoning questions around making the models better. Zoom out comment. It's an amazing business. It grew to $800, $900 million. Think about it kind of one level below the model providers themselves. Th- this is a company selling to those five or six amazing model provider companies a vital service that's probably kind of a pain in the ass if you're OpenAI or Anthropic. You don't want to spend your life rounding up literally thousands of people to answer, you know, what 10 years ago were very mundane questions like, "Is this a stop sign?" And are now, as the AI has gotten smarter, much more advanced questions, as Garrett knows better than me, right at the level of PhD knowledge. So that's the business AI, Scale AI was in, and it was doing $800 million, and its customers were, in the main, the five or six large model providers. That's where we were a week ago. It was a highly valued company, just raised at $14 billion. So that's kind of the, the level set here, and probably the leader in the space in terms of size, not making any comments on quality. And then as you say, a week ago, in the last week, you know, Facebook/Meta announced the fascinating transaction whereby they, that definitely is a bit of a head-scratcher, whereby they invested $14 billion into Scale AI, took 49% non-voting control, allowed the other investors to literally take that money back out as a special dividend, so all those other investors got $14 billion in cold hard cash. Some, and some of the key executives at Scale AI move over to effect- to working with Meta. So that's, those are the facts on what happened, and I think, you know, the open questions that we're starting today is why? What's the impact on Meta? What's the impact on the remaining Scale AI business? And then as Garrett says, what's the impact on all the other providers who probably, as Garrett just hinted, if you think about it, Facebook, again, to state the obvious, Facebook is another contender in the AI model wars. So if you're OpenAI, if you're Anthropic, if you're any of the other guys, you now have a key supplier who's selling you a pretty important subsystem of what it takes to build your model, who is now owned, controlled, dominated, infiltrated, pick your word, by one of your direct competitors. It's got to make everyone pause.

17. HSHarry Stebbings5:22

    Wh- which is exactly what you'll see happen with Windsurf being turned off of Anthropic's model.

18. RORory O’Driscoll5:26

    Agreed.

19. HSHarry Stebbings5:26

    And then back to Garrett. What you're saying is in the last week, the phone's been ringing off the hook.

20. SPSpeaker5:32

    Uh, absolutely. And I think as you talk about the shift from generalist to experts, like, we haven't even talked about, like, the future too. Like, future's gonna involve more audio. The future's gonna involve more tool use. The future's gonna involve more trajectories as some of these agentic systems, uh, and step-by-step problem-solving is trying to be improved by all the frontier labs in their pursuit of AGI, right? So, um, really what that means is that, like, you need experts i- in domain, m- mostly in, in, right now we're, we're focused is on, you know, core science skills, and also skills that you'd imagine, like finance, law-

21. RORory O’Driscoll6:09

    Yep

22. SPSpeaker6:09

    ... medicine, like the large markets that these, that these frontier labs are chasing after.

23. RORory O’Driscoll6:14

    One key question for you though, Garrett. As you're having these conversations-Post the announcement, right? If I was running procurement or ven- for one of the other AI company, and I've had this happen to me once, right? How do you think it will impact how they contract with people like you, companies like you, and how much of the in-- Like, how much of the process will they let you have visibility into? 'Cause one could argue that Scale, the asset that Meta may think it's bought is that Scale AI has a lot of knowledge just by virtue of the questions that are being asked and the kind of expertise that people are seeking from them, has a lot of knowledge about where the most advanced LLM companies are going. And if I was procuring Handshake after that experience, I might have some different perspectives in what I can let you see or not see or... Is there anything you can comment on that?

24. SPSpeaker7:06

    Uh, when we're talking to our customers, I mean, they, uh, we are talking about, uh, the absolute frontier of what's happening.

25. RORory O’Driscoll7:14

    Mm-hmm.

26. SPSpeaker7:15

    Um, and, uh, the, you know, our customers really care about three things that are always in balance. They care about quality first and foremost. You have to have high-quality training data. Uh, you have to have high-quality evaluation set. Like, you, you have to have high quality. Then they care about volume. You know, it's really hard to get to Scale on volumes if you don't have an audience. Like, I would say the only durable moat in, in the entire human data business is access to an audience. Uh, and then they care about speed. How fast can you turn them around? If you're another company in the space, like, you know, or one of our advantages, I would say, is the ability to activate volume and quality quickly. And, uh, we don't have to run month-long advertising campaigns to make that happen.

27. HSHarry Stebbings8:00

    In a year, Garrett, over or under, one word, does Scale have over 100 million or under 100 million of revenue?

28. SPSpeaker8:08

    O- over 100 million.

29. RORory O’Driscoll8:09

    That's a great call, Garrett. And the reason it's a great call, it was a such a savvy thing, 'cause you're probably correct, 'cause it takes a long time to go from 800 to 100. So you didn't diss anyone, and you also gave a probably a factually correct answer. If Harry'd asked a different question, which I'm now gonna ask, do you think the revenue goes down? Do you think customers will reallocate significant spend away from Scale AI to other providers of data in the light of this acquisition? How would you answer that question?

30. SPSpeaker8:39

    What we're seeing right now in the market is that there are hundreds and hundreds of millions of dollars of, uh, spend that is trying to be reallocated to, uh, leading providers. Um, and the primary constraint is the ability to deliver volume and scale, and we c- we can deliver volume, we can deliver quality, and we can do it really fast.

Episode duration: 1:09:30