Ep5. Stock Comp, AI Cold War, Valuations for LLM | BG2 with Bill Gurley, Brad Gerstner, & Bob Mylod

30:15And I want to be very clear like booking holdings, you know, I'm chair of the board of booking holdings and I'm on the comp committee. And I want to be clear like, I view that our employees, I know that our employees have been very, very generously rewarded over, you know, a multi decade period, partly because we live in a competitive marketplace and we have to make competitive stock grants. But the biggest reason is because our stock has accretiated over that period so rapidly. Has gone from $10 a share to $3 ,400 a share. Like that, that is the appreciation that has driven returns for not only the employees, but driven returns for all the shareholders of booking .com.

31:10Correct. And at least from our perspective, our culture has been, hey, the way to, the way to get rich, the way to make a lot of money is for, to create huge shareholder value. And the senior management and the board of directors along the way is going to try to put our thumb in the on the scale in the form of smartly using our capital to further reward shareholders. So bill, again, like you're exactly right. If you look at the top company, certainly the Nasdaq 100 companies, but absolutely tech companies below that market cap, what you will generally see is, you know, a year over year increase in share count, literally the expansion of the number of slices of pie because of these stock granting practices.

32:03You'll see some companies that do share buybacks and they'll literally state I'm doing the share buy back to offset delusion associated with employee, you know, stock based compensation. And thereby doing that, taking a huge cash expense in the form of stock buybacks and putting it below the line so that when people look at operating cash flow or free cash flow, it looks like one number. But yet below the line, there's all this money being spent just to buy back the stock that employees are selling when their stock, you know, vests. We've taken the view that we want to be very, very fair. But again, back to the thumb on the scale, we take our excess cash flow or we've tried to take our excess cash flow and buy our stock back.

32:53So if you were to look at our share count, say 10 years ago, I think we probably had 50, 52 million shares outstanding. And today we have 35 million shares outstanding, meaning, meaningful, meaning that if, you know, during the last 10 years, if the value of our company, the if the enterprise value has doubled, the share price has tripled. And the biggest beneficiaries of that are the people who have gotten the grants that we gave in the prior periods. So to me, creating that virtuous cycle, where you really are counting that stock as an expense, that's the way to do it. And a long, long time ago, we adopted the practice of not performing out the cost of stock based compensation because it's a very, very real cost.

33:47And we do reference, you know, cash flow and we do reference free cash flow. But in our most recent proxy statement, we actually pointed out that the thing that we really look at is, is how much of our free cash flow is being spent on stock based compensation. And maybe that's a segue, Brett, to, you know, how, you know, what are the solutions to get this under control? But like, we have always at booking. And certainly with all the companies that I'm involved with, I try to get everybody to focus on the fact that a dollar of stock based compensation is actually not only should it not be ignored, it's the most single valuable expense.

34:26You had a chart. And in the chart, you compared booking holdings and we'll show this, the percentage of operating cash flow that various tech companies spent on stock based comp. And so again, I've been on comp committees on public companies. So if I'm on the comp committee of a public company board, generally what I do at the start of every year when I'm negotiating the option or the RSU packages for the employees, we bring in a comp consultant, right? And, um, and this comp consultant makes some recommendations to us based upon the peer groups that look like this particular company. Not surprisingly, because everybody has adopted this practice, you know, it's a little bit sh** out.

35:15And so if, you know, if everybody's, you know, following a dumb comp practice, that is your peer group, you know, compare. But tell us a little bit about this chart. And, you know, if I'm, we're on a public company comp, or in them, go, Bill asks you the same thing as it relates to venture. But if I'm on that public company comp board, what is my takeaway in terms of what the gold standard is? What is this the way that I should be measuring it? Or should I be paying attention to compencian these other, uh, comp consultants that are coming my way? I don't know that there's a gold standard. To me, the gold standard is to just count it.

35:51Again, as I said, like you would never think of not counting expenses that are denominated in euros. Um, if you were, if you had a European subsidiary, um, just the the dollars count. I guess what I'm saying is it's like, it shouldn't be excluded to begin with. Now, one easy way of looking at the practices of, of various companies is to look at, okay, what is, what is the dollar value of stock based compensation expressed as a percentage of OCF or FCF? Um, that is a blunt way of looking at it. And by the way, it will shock you when you look at just how much of a company's free cash flow goes directly back to, uh, stock based compensation.

36:39Yeah. I think we've done, we've done that analysis, Bob, you've done that analysis, um, you know, I think on a chart of maybe 40 or 50 names, we have maybe 10 names that are under 20 % of their free cash flow as, as a percentage of stock based compensation. So it is like the thing that jumps off the page at you is just four tech companies, which is what we're, we're, we're talking about what we care about. It's a massive, if not the largest single expense for them. And it's the one thing that way too many of them are not paying attention to. I think to, to give some credit where credit is due, I do think some of the adjustments, for example, that you've seen Meta make, uh, both in terms of headcount and now how they're talking about SBC as a real expense.

37:28Uber's talking about SBC is a real expense like there has partly due to your leadership, partly due to this move to get fit. People are paying attention and are talking about it. And I think from our perspective, it's not that you should stop giving, um, you know, using stock, but to your point, you should discount the damn and treat it as cash and we should have an honest conversation on an all in basis, what these various companies are earning. Yeah. And, and, and look, let me talk first of all, I do think shareholders are starting to care. Like it's, it's, it's, it's, it's, the conversations are happening.

38:06I think most of them oddly happen in one on one conversations. And, and, and as was said earlier by Bob, the industry seems to kind of accept this suggested EBITDA thing. It shows up in the sell side research. It's in the, it's in the earnings releases from the companies. There are many companies who only talk about that on the earnings call and never talk about the net income. So it is, but, but, but I think the shareholders are paying attention. I think they're very unhappy with where things are. A couple of things Bob hit on. You know, if you're in a boardroom where people are saying we should buy back the exact number of shares to offset the SBC, that you were, I'm, this is going to sound like I'm really, you know, but you're dealing with financially ignorant people and they just don't understand.

38:54And those same people, I think, are being widely disrespectful to people in your practice, Brad Abonani shares, you're not that stupid to where you would say, oh, if you buy it back, then it never happened like that. I've had this argument many times, right? You're taking, you're taking my earnings that we could either use to, you know, invest in new projects, invest in new countries, invent it, you know, or, or just hand it, give back by way of dividends. And instead, you're buying back the shares, which effectively just acknowledges that this, in fact, was the cash that we said it was when the, when the R's use vest, they sell, I mean, so you're selling and buying at the same time.

39:37Right. Well, it's a point of that. Anyway, I also agree with, with the comment that was made that I think management teams treat cash. I thought that conversation about the, the hiring example you mentioned about management gets very concerned about spending a dollar of cash. And there, and many of these companies, they almost don't care about spending the equity, which is a weird bizarre thing because they're, they're actually the same thing in these cases. Well, it's not weird and bizarre. If at the end of the year, remember, a lot of their bonuses are based on adjusted EBITDA. Okay. So again, to Bob's point, if we adjust this out, then, you know, it gets back to Charlie Munger, show me the incentives and I'll show you the behavior.

40:20And let me jump off of that. So, so having been in many public comp committee, here's what happens. There is no one in that committee that's thinking about things from a shareholder alignment perspective. They're, they're all thinking about what are the common practices. So you go hire a comp consultant. And the comp consultant's going to tell you what your peers are doing. In fact, you create a peer list, you study what the peers are doing and you build your program based on what the peers are doing. And if, if enough people choose the 75th percentile, then you move, you know, but nowhere does anyone, I don't think the comp consultant or anyone stand up and say, hey, is this a good program for the shareholders?

41:07And when I've tried to do that in the board meetings, I get pushback. Right. And, you know, there's a great, uh, ironic given that, you know, there's a fiduciary responsibility to the shareholders, but I agree with you. No doubt. And, and I love this quote. Warren Buffett said, I'd rather stick a viper down my shirt than hire a comp consultant. And my big takeaway from that is the best and the brightest companies have to have the courage to lead and go in their own direction and create a compensation program that is unique to them and that is financially rigorous and understands these things. We've heard a great example of booking.

41:44I think another one is Netflix. Yeah. So Netflix, today has a remarkably unique compensation structure. No one does anything like them, but today, if you're an employee, you get to choose either options or cash. And there's a slider and you get to pick where you want to be on that slider. They don't have RSUs. None. Yeah. And in a very recent comp committee consultant conversation, I said, I'd rather use options. They said, I wouldn't recommend it. No one else is doing it. Yeah. No one else is doing it. No one else is doing it. And so anyway, this gets back to the conversation you and I had last week, right?

42:21We even have a court criticizing Elon for taking an options package where he made no money unless he saved the company from bankruptcy. Meanwhile, the CEO of his arch rival who created no shareholder value over a period of five years is making tens of millions of dollars a year in our issues. And by the way, I'll say it again, I said it before I would offer any CEO I work with the Elon package and I'm certain none of them would take it. And two, you know, if Delaware doesn't get this right, they really need to understand that shareholder alignment and executive compensation is a very, very important piece of making these markets work.

43:03But I'll just I'll again, Bill, I'll agree with you that because I've been on both sides of it, obviously, when I was on the management side, putting together these compensation plans and presenting them to our compensation committee into our compensation consultant. And of course, as, you know, as a director of many public companies and many, you know, comp committees, the dynamic is kind of almost always the same. A company establishes its peer group. And no board ever wants to think of their management team as being anything other than above the mean. And you have 100 % of all those companies that are thinking of their management teams as being above the mean.

43:47So you have this constantly upward pressure on these on the granting practices. It takes a mindset of like how do we how do we as executives and how do we as leaders of the company want all of us to make money. And, you know, I think one of the back to sort of what are solutions. Comp consultants, comp committees, venture capitalists, even investors, they're very far too focused on, you know, the term, you know, annual delusion percentage, you know, hey, this is a, this is a reasonable granting practice because it's only 2 % or it's only 3 % or whatever. But really all, all anyone is saying when they say that is, they're making a commentary on the valuation of the company.

44:41Because if I'm a company making, you know, a billion dollars of free cash flow, but I have a, I have a 100 billion dollar market cap. Right. And, and I, you know, I've got a 1 % delusion rate. Well, you know, I theoretically look like, you know, I'm managing the company well. It's 1%, even though I just wiped out all my free cash flow. Yeah, you just have to focus on dollars, ignore, you know, delusion percentage. And if you want, yeah, you could look at, you know, stock based compensation expressed as a percentage of free cash flow. But again, Brad, to your point on that slide, it's a shocking number.

45:24The, the percentage of free cash flow that goes right back in to stock is a huge number. And when you do that over and over and over again, if it's 25, 30, 35%, every single year, which, you know, that's kind of the math of it for many, many years, the delusion is actually, if you dilute free cash flow by 30 to 35 % every single year for 20 years, guess what? Your delusion is 35%. You know, it's not, a half of, it's not 50 basis points. It's not 1%, it's not 2%. It's literally, you know, a third of the company. I think the, you know, you make such a good point, Bob. You know, I go back to this Buffett interview he did on CNBC.

46:09And because this gets pretty esoteric. And, and, you know, you sit in these comp committee meetings, I see the eyes glaze over. There's a social dynamic at play. You sit on the board. You're in the room. You often know these people, right? You want to treat them fairly. You know, but Buffett said this and we'll show this, but he said, I think the best way to compensate people is with cash. And if you want to give them a bonus for exceptional performance, give them cash. It's simple. It's straightforward. It's clean. He said, I don't believe in using stock options as a form of compensation. I think it's just a way to get around paying out cash.

46:44And I think it's a way to sort of camouflage what's really going on. Now, I disagree with Buffett on the idea of whether or not we should use options. I actually think there's a tremendous amount of, uh, of, of alignment and incentives that are created through that. But if you ask me, Bob, what do you think is the best thing if you're sitting on a comp committee? The conversation that I would force is let's have a discussion about cash, right? What is the value of what we're giving up? Right? In order to achieve our objectives and does it make sense? Because if not, then it obscures bad business models.

47:27It obscures sloppy performance, right? To me, it comes back to this question of like getting fit. Like you want to know, am I really fit? You know, if you, if you're going to work out and, you know, you're not measuring your weight or your body fat or anything else, you know, like you're just, you're, you're kidding yourself. And so to me, the measure of whether or not a company is really performing is the cash equivalent, you know, um, you know, focus. So Bill, on this, the thing I wanted to ask you is, you know, what has been the rule of thumb in Silicon Valley? Like when, if you're a founder, I mean, you used to invest almost exclusively in series eight companies, first institutional money in and, you know, I don't know, the rule of thumb is there'd be a 20 or a 30 percent, you know, stock option pool in addition to what the founders owned in the business, that would go down over time, I think, and, you know, and as you march to the public market, you would, you would have to focus on, you know, again, treating this as cash and getting, getting profitable.

48:33Where, what are the rules of thumb today in series B, series C, series D? And where do you think, um, where do you think you're seeing some of the sloppiness? Well, look, I don't, I don't, what the thing I'd start with in answering that question is, mostly startups never get to this place that we're talking about, you know, and, and it's a, a lot of the commentary in Silicon Valley is always about the 10 or 20 or 50 companies at the top and the vast majority of these companies never get to that place. And so, you know, the general rules of thumb, as I started with, are like 10 percent headed down to three, it is done as a delusion perspective.

49:14Um, profitability first, I think, is, if you have a company, especially one that burns a lot of cash and if, if new fundraising and capital is expensive, the one thing, I think the one of the reasons people start with this adjusted EBITDA number is it does represent a point in time at which you no longer have to raise capital from the markets. And you could argue over whether that matters or not if you're issuing stock, but it does, it may matter, especially if capital's hard to come by. Um, and then I would just say, you know, some of the same things we've been saying, don't fool yourself, like don't, don't, don't assume cash is cheaper or more expensive than equity, they're really the same thing.

49:55And eventually it's going to matter. Like I do think Wall Street figures it out. I don't think there are any secrets. Um, I think alignment is really, really important. I do this when I'm recruiting too, build, look at scenarios of the stock being up 20%, 40%, 80%, 100 % and look at the paths that people are going to make on that and see if you have alignment. I think if you're only doing zero basis RSUs, you've got a horrible alignment problem. And don't be mad. You like when you think your EBITDA profitable and your stock doesn't have the same multiple as the other company. Um, and, and I guarantee if you look under the hood, they've got a lower SPC expense and they're not, you know, pissing away.

50:42They're free cash flow. Right. Right. So, um, I don't think there are any secrets. I think competitions are real issue. One thing's what I wanted to go to, um, with Bob, you know, I, I, I saw this really interesting presentation from Gary Steele, who turned around Splunk and a similar way to what happened at Meta. Um, and Gary, you know, crushed it at proof point and a huge value creator. And he said, you know, in a very short window, he went, he, he said he took stock, uh, participation from 87 % to 15. And I saw so many CEOs in the audience, like, they're faced, they just went into shock. And it has become almost, um, ritual in Silicon Valley to assume you have to give everybody stock, including the chef and like, you know, it literally is part of, oh, we're cool.

51:32Like everybody else. And so I think that's being reexamined. Like, and then the other thing he said was they moved a lot of employees out of Silicon Valley. Right. And that's an interesting, like thing. And I hear, I'm hearing that from more and more CEOs. They say Silicon Valley is a wonderful place to start a company. It's a horrible place to scale a company. Yeah. My goal is into sort of come on this podcast and say, hey, everybody's getting over paid far from it. It's more just how I think people, you know, we, we need to have a better model by which people get paid. And at least at, um, price line booking, we always had the mantra of, you know, a dollar has to be worth a dollar, whether it's in cash or comp.

52:18And if, if we can't convince somebody that a dollar worth of stock is worth a dollar, then you actually shouldn't grant it. You should just give a dollar of cash because they'll actually value it, you know, greater. Now I think over the over time, because we've been so careful about this and because we have been able to create shareholder value through many ways, mostly through the, through the incredible work of, of the employee base, but also as I said, through smart capital management that has allowed us to reduce our share count, that's an incredible benefit that employees have gotten over the past many years.

53:00Like when the share count is getting reduced, theoretically, the stock price is supposed to go up. And that benefits not only shareholders, but the, the RSU recipient. Hey, Bob, do you have any sense how many employee millionaires that booking price line has made over the last 20 years? Um, no, I don't, but it's, yeah, it is, it is certainly, it is extraordinary number. Yeah, extraordinary number. And, you know, maybe this is a good place, you know, to wrap. I think your leadership on this issue, right? With this fundamental thing, this is not about, uh, undermining founders and employees' opportunity to, uh, you know, make an extraordinary return, make millions of dollars, you know, in these startups.

53:50In fact, it's about protecting that. Yeah. And it starts in the public markets, because if the public markets all pretend that this doesn't exist, then they're going to pay excessive amounts, which deprives startups and the venture capital community of being able to hire those employees and compete. You know, let's see, example, you just gave us, uh, the, the, just now of Microsoft. And so I think that accountability, treating stock as cash as public shareholders and the good news is, like I said, there's a lot of evidence that this is starting to, you know, uh, you know, be a real matter of conversation on Twitter, on CMBC, in annual letters, you know, folks like Bob leading the way.

54:33And then I think we need to, uh, you know, begin telling the stories again in Silicon Valley, the reason the Michael Dells of the world own as much of their companies at the end of the day as they do is because from day one, they were concerned about giving shares away where they didn't need to give shares away. And so that discipline cascades down. I think all companies are better when they're DNA from the get go, right? It's just honest, right? They treat all of these things as expenses. It forces you to build a better business model, right? We were lucky to be partnered with Bob and investors from the early days.

55:12And they were forced to really reboot and think about the business model, uh, that they were going to focus on. And if they kept pretending, right, that this didn't matter, I think it would have been a very different outcome for that business. I think you would have produced far fewer millionaires than the number that you've produced. And, you know, again, in Silicon Valley today, I do see this also cascading down into the companies that we're investing in. Certainly the companies that were on the comp committees of the boardrooms that we're in. This is a matter of conversation. Um, you know, I would say during the heyday of Zerp, um, I was seeing, you know, some really crazy stuff in terms of annual dilution.

55:53And now it's a tough conversation at each of those boards because I think a lot of venture capital firms and founders woke up and said, wow, I just gave away a lot of the business. I thought that was okay because the valuations had gone from 500 million to 5 billion. But now that I know I'm really only worth a billion dollars, you know, the consequence of that dilution was way more dramatic than I thought. And I would just close by, by, you know, once again, going back to this notion of independent leadership, the only companies I see kind of getting out of this box are ones where you have people that are thinking from a first principles standpoint, like Bob did at booking, like, you know, in the Netflix example I gave, you know, they're only diluting 1 .5 % a year, which is pretty, uh, left rail kind of thing if you look at them on one of these charts we're going to post.

56:45And so, you know, it, it, it takes someone that really understands the stuff at a fundamental level to be willing to step outside the box because once again, the Compton Sultans, you know, maybe even ISS are going to tell you to do whatever the else is doing. And, um, and I don't think that's going to get you to the finish line. Right. Thanks for being with us, Bob. Bob. Well, Brad, I thought maybe we'd hit, uh, two or three more topics today and, and, and wrapping it up, I just thank you to once again to Bob and I wanted everyone to know that we're going to put a bunch of resources in the show notes.

57:19We've taken, uh, snapshots of 50, 100 different companies looked at SBC as a percentage of free cash flow, which, which Brad and his team believe is the most important, but also revenue and market cap. I think those are all useful ways to kind of get a sense of how companies are doing on this issue. And then lastly, um, my good friend Mike Mobison, who many people in the internet finance world know about, he, he put out a really great piece on stock -based comp and, and, you know, reading a, put a lot of thoughts in my head as, as we went through this, so I'll put a pointer out there to that as well.

57:56Okay. So there's some, a lot of new stuff happening, Brad, um, since we talked last. Um, one of them that's kind of interesting and I think, if I think I saw a photo on the internet of, uh, Sundar and Tim Cook at a restaurant that may have kicked this off, but there's a rumor that Apple's talking to Google about leveraging Jim and I, I believe. And tell me what you think of this. Yeah. I mean, I, I think the news broke yesterday. Both our stocks were up a bunch on it. This idea that, um, I think the implication of the rumor or the, the, the tweet was that Apple was somehow maybe outsourcing or partnering with Google on Gemini.

58:36And, um, you know, I tweeted, uh, in response to this, that it, you know, I put the probability of them, quote unquote, outsourcing AI to Gemini at next to zero. We talked about this on the pod a few weeks ago. And I think that we, you know, I go back to that analysis, which is this. Um, Apple is in this enviable position to be able to build, right, the killer personal assistant. Yeah. Um, that has memory that has persistent that can do transactional thing for us, whether it's make a phone call, send a text, book a restaurant reservation, book a hotel, et cetera. So we suggested that they're going to work on their own native small language model that really doesn't need to be Einstein.

59:23It just needs to serve me really well as my personal assistant. I think what's likely, you know, being commingled here is I presuppose that Apple is also talking to Google, to meta, to open AI and others about bringing those models in, right, as part of a generative search application that will sit on the deck as well. So think of that maybe similar to perplexity. And then of course they have their search deal with Google. So, you know, to me, I don't think anything really changes in my view about Apple. I think they continue to be in this really interesting position. I'm sure they're going to talk about it in June at their developer's conference.

1:00:07Um, I think it's unlikely that you're going to see some transformative thing get, get shipped this fall. I think you may very well see a generative search application that gets searched this fall, get shipped this fall that may in fact have multiple of these answer engines embedded inside it, including Gemini. But you know, the final thing, you know, I would just say on this is, you know, I've been over the course of the last couple months pretty openly uh, uh, uh, uh, critical in my critique that Google needs to face up to this innovator's dilemma. Yes. Um, that needs to make some real changes in terms of how they're going to market.

1:00:47And, you know, I, you and I have those debates all the time. And I, I think they're, you know, is a group of people, you know, on Twitter and otherwise, who assume that I must somehow, you know, and we've said many times this show is not investment advice. Um, don't presuppose, you know, you know, where we stand on a particular stock. I'm often asking tough questions of stocks that we own. I mean, go back to, to meta. Um, and in fact, we own Google, we weren't shorted this week. You know, what, when, when this news came out. And so I think, you know, it's an important reminder, you know, much like Stan Druckenmiller.

1:01:30At this moment in time when you're in the middle of these phase shifts, mental flexibility is incredibly important. And, you know, we are going to follow the facts and we may change our opinions on a particular stock. I think ultimately for Google, you know, the challenge remains the same. The challenge for Apple remains the same. You should imagine that everybody is talking to everybody. Yeah. Right now is they try to source, you know, sort out where they're going to, where they're going to compete. But I don't know. What was your reaction to it, Bill? Well, it's funny because you, I think you can have two, two reactions that are exactly opposed to one another.

1:02:07So you could view it, you said both stocks were up. So, so that's the interpretation that this is positive for Apple that they're, they're maybe getting their AI game ready to go. You could view it as weakness on Apple that in order to get AI right, they need to go talk to Google and they can't do it on their own. As we have said before, I think the handset is a remarkable asset in the long term AI race. And, you know, there are one issue that could come out of this is regulatory. I mean, if you combine the two of them, that's nearly every handset. So if the same AI code we're running on both, I don't think that'd be great.

1:02:46I do believe that the handset is so valuable here. If I were Apple and wanted to prove to the world that they had their AI game together, and I know this is a bit of a broken record with people, but fix Siri. Like Siri evolved and released. But that is the small language model that I'm talking about, right? That is going to be the new back end for Siri. Like, there's no way they're outsourcing that to them. And by the way, on that topic, I think voice AI may be more interesting than say Sora or the video stuff that opening I released. I think more and more people are realizing because LLMs are language models that people would love to just talk to them.

1:03:27You brought up the point that Meta may be more focused on the sunglasses with an earpiece, right? Right. Right. And you want to be able to talk to that. And so for me, the irony of this is I think voice AI may be a new frontier that you may see a bunch of these people race after. And getting that to work appropriately so that you feel comfortable, always like to reference the movie her. Just talking to this thing and not having any errors. I think that's super interesting. So we'll see how this all plays out. Let's move on to the next topic. So there's been a lot of conversation in the past few weeks about TikTok and bite nants.

1:04:10And obviously it very quickly moved to the center of almost all discussion because it involved the house and the senate and it's on the Sunday morning talk shows it's in every newspaper. And it's been discussed quite a bit, but I'd love to hear your thoughts Brad on this situation. Yeah, you know, I've said publicly before, you know, we were early shareholders in bite dance. I think we've been involved almost a decade, right, in the company. So take everything I share with that grain of salt. First, I think, you know, everybody knows bite dance is a super large company. There's a lot of numbers that are rumored out there, but the reporting suggests that over 90 % of its revenues are outside the United States and over 100 % of its profits are outside the United States.

1:05:00What's that mean? Well, I think I've seen other numbers out there that they're doing somewhere between eight and 10 billion in TikTok, US revenue and that they're losing money on TikTok, US. So I think it's important just to understand where that sits in the context of the entire business. Secondly, with respect to myself, you know, I'm a dad and certainly an American first. And, you know, my kids use TikTok, they create on TikTok, etc. But, you know, I, you know, I read all the same things everybody else reads and I think it's impossible for me to prove or disprove. And I think there's legitimate debate as to what's going on.

1:05:36You know, I think there's also lots of legitimate debate about the house bill itself and whether or not that's narrowly tailored enough, whether the language needs to be improved, etc. But clearly it passed the house, you know, by a wide majority. I think there's a real question as to whether or not that will pass the Senate. But I think that misses the point a little bit. Set all of that aside. What's become clear to me over the course of the past 12 months and certainly over the course of the past 30 days is irrespective of the merits of the matter, right? This is now in the sovereign domain, right?

1:06:17AI over the course of the last 12 months has gone from an interesting technology to literally the leading edge of where we're battling with global sovereigns over trade, national security, national economic interests. So given that, I think it's almost inconceivable for me now to see how TikTok US has a path forward under kind of the status quo. So where does that leave us? It leads us on the one hand with, you know, the proposal that I think, you know, President Trump had made toward the end of his term. And which I hear bantered about by lots of senators and House members, which is some sort of spin to US shareholders, where maybe bite -dance swaps some of its ownership or sells some of its ownership to somebody like a Microsoft or an Oracle.

1:07:09And the app can continue running in the United States. Okay. I think there are a lot of US regulators that would find that an acceptable outcome. Yeah. But I'm increasingly of the mind that the Chinese regulators would not find that an acceptable outcome. Okay. And the US regulators won't allow the app to continue as is. So I think there's a real chance that when we're looking at this 12 -day -tea months from now that it's an absolute no -go that they can't find a middle ground and that the app is shut down. Which, of course, we saw President Trump tweet like the implications of what that makes Meta more valuable or more powerful as he suggests or perhaps these other US companies.

1:07:52But the thing I guess that interests me the most is what is, you know, that's just the first step bill, right? Like that's the first the first move on the AI chess board between these two sovereigns, if you will. It's hard for me to see a world where TikTok US is shut down or banned where there's not some response by China. And I think we just had this conversation about Apple putting Gemini AI on the Apple phone. I'm certain, you know, Apple China is a monstrous business, is a huge business in China. And I imagine that now that AI is going to be embedded in these phones that there's probably a little bit different conversation that's being had in China about, you know, about that there as well.

1:08:37So I think we're seeing just the early innings, unfortunately, right? I tend to think that global trade is a good thing for human prosperity. I think it's driven a lot of our global GDP and prosperity over the course of last 20 years. So I don't celebrate that we're entering what appears to be a global AI cold war on the economic front. And I'm certainly not critical of the fact that US regulators have these concerns. But unfortunately, I don't see much of a way out here. I hope I'm wrong. I hope that, you know, you know, when President Xi was here a few months ago, he said, I don't want a cold war, I don't want to hop war.

1:09:17I want a partnership. But that's not what it looks like from my vantage point today. Yeah. I'll give you a couple of my thoughts. First of all, I thought our friends over at All -In did a really great job of tearing this apart. And even where they disagreed, I think every single one of them agreed that the data probably should not be going to the CCP. Like I haven't seen anyone said, oh, I'm completely cool with data going to the CCP. So if everyone's in agreement on that, and I do understand the arguments that maybe the House bill is too flexible or too vague or too broad, and you could tighten it up and achieve the same things.

1:09:54As I've told you, I feel like that the company, and just the company I'll say is either US TikTok or by dance, could have done a better job of proving data independence. For sure. And I think they could have hired an auditor, you know, a KPMG or a price water house or whatever, rather than simply saying, oh, whoa, it's in Oracle and Texas. Like to me, there's not much certainty in that comment. I would also say relative to your Apple comment, I do believe Apple has to keep all their data locally over there and prove it to the CCP. But that may not be enough. No, it may not be. But I'm just saying from a repress, prosody standpoint, I think they're already doing what is being asked here.

1:10:38And as we all know, and we've said the social networks here aren't allowed to operate there. So you could have a, I think you have a pro -China mindset and just still hold the line and say fairness is fairness. So we will see how it plays out. I'm hopeful that there's a way to get it together. But who knows? We will see anything else on that? No, I mean, I think that it's once you start going down this slippery slope, right? You know, we have a lot of US companies that have a big footprint in China. And so we're already down the slippery slope. Well, I mean, we haven't banned it yet. So I think if and when this effectively, you know, if this were to get banned in the United States again, I would like to think that that would be, you know, where this ends.

1:11:30All I'm suggesting is that chat GPT only came out 18 months ago. And you can't use the China. Well, the wheels of international relations, I think, turn slowly. And when we look at its first started with silicon, right, with restraints on China's ability to get Nvidia chips. And again, I don't disagree with, you know, our ability to keep our best stuff to ourselves. But it's hard to look at the facts on the field and say that this is not, we're not entering a new phase of global international trade relations, you know, vis -vis China. Now, one might say that we're just, you know, creating a more fair and balanced plane field, right, relative to the restrictions that have been imposed on US companies.

1:12:20But I think it would be naive of us to think that there's not going to be, you know, counter responses out of China. I hope not. I take the other side of it simply because we're not even getting to the point of equality here. So anyway, we'll leave it at that. And then this morning, just this morning, we rarely have news that broke just this morning. A very unusual thing has played out with Microsoft and inflection. Why don't you tell people what went down? Well, inflection, you know, is, you know, just as a reminder, was one of the early LLMs to be trained on H -100s. They have a consumer product called Pi, with stands for personal intelligence, founded by Mustafa Soluman, one of the co -founders of DeepMind.

1:13:08You know, and in a friend. And, you know, what they did was, you know, as one of the companies. By the way, just to put it in, I think there's another way to frame this. There were three LLM companies in Silicon Valley that erased over a billion dollars, although some of that was with credits, but there were three and Thropic, OpenAI, and this one. Correct. So, you know, there's a large startup, but what they were going after was really think the chat GPT consumer market with Pi. And, you know, today we heard news that Microsoft has effectively hired the employees out of inflection. We don't know all the details, you know, of the deal here.

1:13:56But, you know, we looked at investing inflection as many others did, and, you know, one of the reasons that strategic investors have displaced a lot of venture capital investors in the LLM game, and we talked about this, I think, on, you know, episode two, is the quantum of capital it takes to compete in the LLM space is very high. And on the other side, understanding the durability of the business model, like in this case, getting consumers to pay you, you know, for Pi is far from clear, particularly in a world where you got to go compete against Apple, you got to compete against Meta and others who are going to build a personal assistant.

1:14:39And so, I think it was a very difficult venture bet to underwrite. I think it was a logical bet for folks like Microsoft, you know, to underwrite because, you know, they want to be in the consumer, answer engine space, the consumer chatbot space, if you will. I think the most interesting thing out of this to me, and if you read such as Tweet that he sent out this afternoon, is he said he's excited to have them build consumer AI, which, you know, he said the co -pilot, like the co -pilot, that is loved and benefits people from around the world. So, just step back from a second here, he owns 49 % of open AI.

1:15:18Chat GPT is charging really hard to be, you know, kind of my personal co -pilot, if you will. And now they have another, you know, another team in the hunt. So, you know, I think this is smart by Sachin, Microsoft. Some might view it as hedging their bets a little bit against Chat GPT, but I think it's great for the consumer. They're going to be a lot of at -scale players that are vying to be your personal assistant, that are vying to answer your questions. We just talked about Apple. We talked about Met AI. Google's certainly going to be in that hunt. And now it looks like Microsoft is doubling down on the consumer co -pilot, which will bring even more competition to the table.

1:16:01What was your take away? Well, look, I mean, I got to tell you, I was floored a bit when I saw this, mainly because of peculiarity of it all. So to have a company that is one of the three, let's just call it biggest backed AI companies in Silicon Valley, not get bought. Like if they got bought, that means, oh, Microsoft thought this was strategic. Microsoft, you remember when Facebook brought what's that? It was 10 % of market cap. So 10 % of Microsoft's market cap is 300 billion. So they could have paid 10 billion. And the last round here was at 4 billion. And you'd go, oh, great. Everyone loves AI.

1:16:47Investors are getting paid. Everything worked out. But this isn't everything worked out. This is employees. All the key employees leaving the company. And the blog post suggests the company goes on with a new CEO. And to me, I just had to look this up because I don't trust racing. But the yellow flag comes out when there's a hazard on the track and everyone needs to slow down. And you look at this series A here, a well -known venture firm, GrayLock, 260 or 250 at a billion pre. These are these AI deals that have remained like Zerplike. Right? And this is a warning sign that you can't just pay any price for any deal in the AI sector and get paid.

1:17:35And because I don't, you know, this, I don't see this now. I don't see the current status as a company that's going to have a venture like returned from here. There's a lot of details. There's a lot of these credit things that are all tied up. There's stuff we don't know. But it's unusual. And I view it as a yellow flag that we should slow down on the track, on the AI track. And maybe be a little more careful. One of the things that it brings up too is a conversation you and I were having, just about LLMs generally. I spent the last two days down it in video, GTC, their Global Developers Conference.

1:18:15It's just extraordinary. The ecosystem of AI that they are consuming. This is not a chips company. This is not just a systems company or a super computer company. They are now expanding on the CUDA stack with a totally new inference software cluster called NIMS. And NIMS, you know, they have embedded, you know, open source models that are, you know, highly and tightly integrated with their overall software stack. And one of the things we've talked about and debated the last 12 months is just the durability. Like what's the business model for the LLM? Yeah. Is it going to, you know, you have these open source models by well -financed companies like Meta that don't seem intent on charging for them?

1:19:06And when you have companies that are valued at $10, $20, $90 billion, right, around that, the real question was about the durability of revenue. And, you know, we had a, we have a vibrant debate within Altimeter on this, right? So the the acronym we all know when it comes to software companies like these LLMs is ARR. So everybody wants to know what's their revenue. ARR stands for annual recurring revenue. And recurring revenue like a monopoly cable subscription gets a really high multiple. That's why softwares have gotten high multiples. They're very predictable. But when we look at what's going on in AI, I, I've banned the use of the term ARR within Altimeter when we talk about AI revenues because they're certainly not recurring like.

1:20:01It's option based. And they have some amount of coaxide. Right. And so one of the, we call them ERR, experimental run rate revenues, right? And so when I look at these, you know, one of the things that inflection and these other models that it's been very challenging for us to get our arms around backing, even though they have extraordinary founders and are building extraordinary things is just if you have to pay a high price, then you need to have confidence in the durability of those revenues and profits on a go -forward basis. And I think you have some thoughts on this. Just look, there's nothing that I spent more time on recently just trying to learn and ask questions.

1:20:45And, but, you know, both trying to understand how these things work, but also how they're being used. And one thing you constantly hear from developers on LLIMM, and let's separate this maybe from the consumer game that I guess, nah, and then there were two or three, right? You got Microsoft. I guess you got Google perplexing, opening up $20 a month. That's a whole separate conversation. But on this B2B side, the model that just seems so prevalent with all the companies I talked to, you might test your thesis against one of the premier models. But then the minute you go to run it, you then run it on four or five others and see which one and different price points and see which one is cost effective for you.

1:21:28And the per misguity is unbelievably high. Another way of saying that one of the things that leads to high valuation multiples of switching costs and today, you know, as it stands, I would say if you were to rate these LLIMMs on their switching cost with zero maybe being none and 100 being perfect, I'd say they're two. Like it's remarkably low. And there's another huge irony. Some people believe, and I won't try and make this argument. I'll just say I know some smart people have said it, that as the context window gets bigger, they'll be less training and even less fine tuning, which means you're just dumping data into the prompt.

1:22:12And this is, this data is not being dumped in in a structured way. It's just PDFs or CSV files or whatever. There is no lock -in. You just dump it in the next one. So I think one thing that's imperative on any of these LLIM models, if they want to be successful in the long run in terms of creating equity value, they've got to find a way to get the switching cost. And I, you know, I don't know what that is exactly right now. I think some element of fine tuning or embedding data or memory like we've talked about, but I don't think they have them today. And maybe this is, maybe this is a sign of that.

1:22:44I don't know. It'll be interesting to watch. Well, I think along those lines, I had dinner this week with one of the heads of data, an AI at one of the largest banks in the United States. And they did what I think a lot of other people did after they saw ChatGPT. So remember, ChatGPT comes out in Q4 of 22. The CEOs of all these banks and every other enterprise went home and that's what the conversation was around the holiday table. So the CEOs or the board came back in Q1 and they said, hey, what are we doing in AI? And they turned to their tech team and they said, let's get something done. So in Q2 of 2023, you saw all of these new customers sign up the Azure OpenAI API, right?

1:23:27Or do it directly with OpenAI because it was the only game in town. But quickly, all of the other data platforms, whether you were AWS, whether you're Google, Snowflake, Databricks, etc. They all spun up their own AI capabilities. Now they all embed free models, whether Islam, Mastral, etc. Even Microsoft now has competitive open models. And my hunch is that OpenAI will also be found on some of these data platforms, not just on Microsoft. And so it seems to me the switching cost is not with the LLM. I agree with you on this. The switching cost is really around where your data resides. Right? We've heard this talked about data gravity.

1:24:14So this particular bank, which had spun up the OpenAI API in Q2, has been playing with it on a big rag over the course of the last year. But I asked them what their plans were going forward. They said, oh, we're moving, we're going to run all of our AI in AWS, right, on open models on our production data, right? So it seems to me that the AI is finding its way back to the data. And the only way, I think for Azure, anybody else to keep this experimental revenue when people run to use these models is you have to deliver something sufficiently differentiated that it gets somebody to pull all of their data to you.

1:24:55And that's a hard road to home. You know, and I'm reminded, I don't know if I've used this metaphor before, but there was a time before digital music where you just could, there was so much music in the world, you couldn't imagine it. And I remember one day back in the Napster days where someone tapped me on the shoulder and you said, you see this and he was holding the hard drive and he said, this is all the music I've recorded. And like he just had it. And to a certain extent, you know, the LLM, it may be, and I've heard this argument from some academics, it may be that most of the data think what could be to everything we know is already in there.

1:25:32And like that it's gotten in, you know, small very quickly. And what the reason I bring that up relative to what you said is, is it may be that the open source LLM's are good enough. And that putting them approximate to your data is a smarter play than trying to put your data approximate to the leading proprietary LLM. And if that plays out, that'll be super interesting. And one thing's for sure, the open source LLM horses out of the barn. Right. It's running wild and free. I mean, and you got to, you got to love the competition. And, you know, again, it's a, you know, you and I have talked about some of the concerns about regulatory capture as, you know, and preventing this competition, you know, going back to the conversation about China.

1:26:24It's so important that we protect this competition among these models because it's driving down the cost to all enterprises. It's driving down the cost to all inference to consumers. And, you know, all I know is it's going to be an exciting few months ahead. So always fun to have you here. Fun to do this again. All right, man. And until next time, take care.