Transcription
All of these companies, Microsoft, Meta, Alphabet, you name it, they now have to go use their balance sheet to a degree they wouldn't have had to otherwise, technically. And so, I have no problem with that, run your business the way you want to, but present it a little more cleanly so people understand, well, okay, part of this is not just AI data center build out or whatever it is, part of this is compensation. And as an investor looking to value the business, I need to know what the real number is that I should use.
>> [music] >> Hi, I'm Telis Demos. Today's show we're going to be talking all about the Knicks. We're going to go over every basket. No, I'm kidding. Actually, today's show is about AI hyperscalers and free cash flow, but don't worry, it will be just as exciting as all the NBA playoff games were.
Uh first, I want to introduce Kevin Coharki. Kevin, say hi.
>> How you doing, guys? A real pleasure to be here. Thanks for having me.
>> Kevin is a long-time financial analyst and a principal at CAE Consulting. He's also a professor at Purdue University. And over here we've got the WSJ's own Jonathan Weil. John, say hi.
>> Hi, Telis. Hi, Kevin.
>> Uh John works with me at Heard on the Street, and for those of you who aren't already familiar with John's work, he is basically the goat of accounting-driven reporting. Uh and he's been doing it since the days of Enron and WorldCom.
Uh all right, so why why have I got these guys here today? I know that a lot of you out there in the audience are very worried about whether or not we're in some sort of AI-driven bubble. And I know that that worry has become especially acute after SpaceX's IPO and its blast off now as of when we're recording toward a $3 trillion valuation. So, as we get into the start of earnings season, we wanted to provide everybody with some ways to think about whether or not the results of these companies are in any way matching or keeping up with the hype that we are seeing in the pricing and just conversation around them. And that's companies like Nvidia, Microsoft, Alphabet, Meta. And along those lines, I think that something that John Wylie has been writing about has been a really good guide to how to think about some of that. And I know John some of that story started with a conversation with Kevin. So, why don't you take us behind the scenes of of how you guys first came into contact? What conversation did you start having coming out of that?
>> I think it was February and you just cold called me or emailed me.
>> Pretty much, yeah.
>> out of the blue and said I've got a story for you about companies that are completely overstating their earnings, aren't profitable even or barely profitable and it's all because they understood the costs of stock-based compensation and I want to show you how.
That's where it got really interesting because then you can look at you know, Meta was the example that we that I used for the column in February. You could see that well, they start off with X amount of net income and it looks really impressive. And then you look at all their capital, you look at their cash flow, it still looks from it's called a standard number called cash flow from operating activities. It still looks impressive. Then you look at how much their capital investments are, CapEx it's often called. Then you say, "Oh, it looks less impressive. They're spending a lot of money on data centers." And then you have to take into account the cash costs of stock-based compensation which basically consists of paying taxes, withholding taxes for employees, and then buybacks that are related to the actual stock-based payments themselves because you're trying to offset dilution to keep shareholders from having their stakes diluted. You take those two elements and now you're down to like almost no earnings, almost no free cash flow. And the free cash flow again, it's it's what is left over to either pay down debt or you know, send cash back to your shareholders once all of your capital investment and all your compensation activities are done with. And at Meta it removed like for 2025 it removed like 96% of their uh what, free cash flow or cash flow from operating activities? Which one?
>> I I always put it in cash from operating activities cuz to me it's it's compensation. So, whether I'm paying uh you and John in cash or I'm paying you in stock-based comp, makes no difference. If I If at the end of the day I'm going to buy back the stock and pay cash, then that to me is an operating activity. So, the adjustment I make is I actually reduce operating cash flow. And then to John's point, take out capital expenditures and what you're really left is what the owner would walk away with, which is the which is the adjusted free cash flow that I think John and I like to like to call it.
>> So, backing up a little bit, we know that it cost an enormous amount of money to build a data center, right? Buying that stuff would be reflected in a company's capex, right?
>> Mhm.
>> And then of course, you need people to come up with models and test those models and do all of the engineering that goes into, you know, developing artificial intelligence and then thinking of products and those people are highly sought after and you need to pay those people a ton a ton of money. And not all of that pay is in the form of salary, some of it is in the form of options that they can then exercise to get stock and make tons of money that way.
>> Or restricted stock, too.
>> Yeah, restricted stock.
>> Or or restricted stock.
>> Okay, but you're you're offering them some sort of stock-based compensation to work for you. And that's really important, right? Like these people have lots of options. They could work at Anthropic or Open AI or Meta and they're highly sought after and so to to recruit them, you need to offer them really big pay packages, right?
>> Mhm.
>> Pretty much. And it gets undercounted on the income statement.
>> Correct.
>> Yeah.
>> So, John, you were trying to get to the heart of how much money essentially, the free cash flow. What what what actual money is sort of flowing in or out of these companies on a quarterly basis. When you and Kevin started talking, um what what conclusion did you ultimately come to? What what is the the the right way to think about the free cash flow of these businesses?
>> There's one number on there that was you know, not in doubt. It's like, you have tax withholding costs uh related to these stock option I mean, you know, restricted stock grants. Uh that's just disclosed straight up for the vast majority companies. That's not hard. The hard part is when you look at their stock buybacks, trying to come up with some reasonable estimate of how much of those are related to actually paying the employees versus buybacks for other reasons. Like, we think the stock is cheap or we just want to return capital to shareholders or whatever.
>> kind of the regular buybacks that companies do all the time.
>> And so, it it's probably too complicated to go into here. I'll let Kevin give it a shot later if he wants. But, uh to, you know, try to estimate what percentage of buybacks are uh compensation related. And some companies' disclosures are much better than others. Meta actually has very good disclosures on this because they give very granular data on a part of the financial statements called the equity statement. Uh statement of stockholders' equity. And based upon the disclosures in there, uh we were able to estimate that 90% or Kevin was able to show me how to estimate that 90% of the buybacks for last year were related to uh stock-based pay. And basically, you can see that that 90% uh the whole purpose of that was to uh offset dilution. So, that ordinary shareholders wouldn't have their stakes diluted.
>> Backing up for a second. So, that dilution is So, if your employees are exercising options or using restricted stock units to get actual stock.
>> Yep.
>> Right? The stuff that they can actually then take and sell and get their money out. If they if if all of your employees start doing that, basically that creates a lot more of your stock, right? And as an outside investor in the company, you're thinking, "Wait a second. The company's handing out all these shares."
>> Yep.
>> The more shares there are, unless the value of the company also goes up, which in this case it isn't really, then I'm just getting a smaller stake of the company because you're just handing out, right? And as an outside investor, you're like, that's a problem. And so what the companies did in response is say, "Don't worry, we will buy back the stock that that that we are creating to give to employees." But then that money they were spending wasn't then getting reflected in the free cash flow.
>> Correct. Yeah, cuz the way the free cash flow is normally calculated, you just have cash from operating activities minus capital expenditures. I'm a Buffettologist at heart. So Buffett and Munger said even after they started expensing stock-based comp, they said, "Now we adjust for the true cost." And so for years I always thought, "What do they mean by the true cost?" And as just as my accounting knowledge got better and whatnot, and I thought about this more, I realized, "Oh, I see what they're doing." And and it's very simple to me. If we were to take Meta off the stock exchange, if the three of us were to get together and buy Meta, we would say based on 2025 numbers, they generated 45 billion roughly in free cash flow. And so let's say we're going to pay 20 times free cash flow for that. We're going to pay 900 billion for this company. Well, by when you do the math that John and I do, we would be very disappointed at the end of the year because we're going to find out that it's not even 5 billion in actual free cash flow because you have to adjust for the stock-based compensation. And then there's a $2.5 billion adjustment for the principal portion of finance leases, which they call out. They say to do that one. They ignore the 800-lb gorilla in the room. And so I look at this and I say, "I've had people argue with me on this, investors that manage billions and whatnot." And I've told them, I said, "Think of it this way. If the three of us were to buy Meta, take it out, those engineers that you had mentioned earlier tell us, "We're not giving them stock anymore." So in order to keep them, we're going to have to pay them in cash salary or bonus, or they're going to walk. And as soon as we pay them in cash, now it's on our P&L as SG&A, selling, general, administrative.
>> in earnings.
>> Correct. And the net income goes down. Okay.
>> And so does the cash from operations.
>> And free cash flow, unlike uh things like net income, it doesn't have a standard definition. But, by and large, uh the way most people calculate it is cash flow from operating activities, which has a standard definition, minus capital expenditures or CapEx, again standard definition, and the just the vast majority of people have never thought through to exclude all the other actual operating costs that aren't in the operating section of the cash flow statement.
>> So, how did we get to this point, Kevin, where this absolutely necessary cost of hiring the best people to do this work was somehow not being factored into a company's free cash flow in a way that that like you guys have laid out really makes sense. Why wasn't that always part of the calculation?
>> So, the way the way I think about it, it's it's two kind of accounting rules. We have very basic, the matching principle. So, any any time an employee works for you, you can match their expense to the revenue it created in that period. We do the same thing with buildings. We might depreciate a building over 30 years. And let's say the building costs $30 million. Okay, you're going to depreciate a million dollars a year. Why don't we just dump that all into the income statement in year one? Well, because we can match the expense of that building to the revenue it's going to generate over a 30-year period. So then, when you get to stock-based compensation, it kind of works similarly to where most of the time these things will have a three- or four-year vesting period. Now, some of them cliff vest all at once. Some some will do evenly over three, four years. It doesn't matter. To John's point, let's say that we have an option or an RSU at 30 at the grant date. That's the price. And it's if it's a three-year vest, you can't touch it. You can't do anything until it vests. So, we expense $10 a year for three years. Now, the life of that thing could be 10 years. So, we have no idea. And so, because we don't know that, the FASB, in my opinion, basically said, "Well, let's be conservative."
>> Correct. The Financial Accounting Standards Board, right. They set the rules.
>> And RSUs is restricted stock units, right?
>> Right. They say, "All right." Cuz I believe this is how they do it. They say, "All right, well, we're going to be conservative. We're going to expense what we can now, follow the matching principle." And then when Telus or John go exercise or the option or the RSU vests and the stock is at a at a 100, fine, that is what it is. But that extra $70 doesn't go on the P&L ever. And I'm sitting there looking at this thinking, I understand why FASB does it. I I don't disagree with that. And so to the way John wrote the article and whatnot, we're not saying anybody's doing anything illegal far from it. But it's just a good analyst needs to know how to think about this and make the adjustments. And as FASB flat-out says, I'm going to paraphrase, of course, but the financial statements are meant for for lack of a better word, quote, knowledgeable users, end quote. And so if you don't know how to think about these things, well, that's on you.
>> So I know you guys have both prepared spreadsheets for our our episode here because that's that's that's just sort of
>> we do.
>> the type of people you are. So, let's talk about magnitude here. We've been talking a bunch about Meta. I know it's not exclusive to to Meta, but the that was one company you guys looked at. What are we talking here, magnitude? What would a an untrained investor look at and say, "Oh, here's what their free cash flow is." And what would you guys say is the right kind of true number that people should be thinking about?
>> So when I looked at Meta for 2025, it's 89% of their free cash flow goes away. And the difference, John, I think your number's a little higher because
>> Cuz I was using an exclusion that they themselves do, the finance leases.
>> Yeah, they Meta's one of the few companies that actually include an exclusion. It So me and John's math actually is the same. I just didn't include it in the number I just gave you. My calc for Q1, it was 33% because Meta stopped their buybacks. Alphabet's the other one. Alphabet for fiscal 25 was 67% free cash flow goes down. For Q1, they also stopped doing their buybacks, but it was still a 62% hit.
>> Can you give me just some raw numbers on that? We're we're talking what billion to what billion?
>> For Meta, it went for 2025, free cash flow went from 46 billion to 5 billion.
>> Okay. It's a big It's a big number.
>> For Alphabet, it was 73 billion free cash flow down to 24 billion. Uh for Q1 for Meta, it was 13 down to 9. And for Q1 for Alphabet, it was 10 down to about 4.
>> John, does that check out?
>> Yeah. Uh Meta, I would actually take it down a little bit more because of the one other adjustment we talked about.
>> And of course, there's, you know, always multiple ways of looking at something. So, what would Meta or company like that say to the to the findings you've been talking about here?
>> Well, for that particular column back in February, Meta declined to comment. So, but there was never any disagreement about the numbers. I mean, the numbers are what the numbers are. The question comes down to, you know, going forward, can they generate the cash necessary to keep on financing the buildout that they need to do or that they believe they need to do to be competitive? Um and they would say, "We'll find a way."
>> All right, hold that thought. We're going to take a quick break. We're going to let everybody breathe. Now that you've had a had an accounting lesson here. Uh take a break, go take a walk around the block, come back, and we're going to talk about why this matters cuz it really does get to the heart of everything that we've been talking about with the AI, you know, buildout and the enormous costs that are associated with that. So, when we come back, we're going to continue with Kevin Coharki and John Wyld.
All right, welcome back. So, okay, we've been talking about free cash flow at big AI hyperscaler companies and why the number that you see, if you know, if you're not that experienced and you need to go look it up, why that number might not really be the right number to think about. And I want to ask you guys, Kevin, I'll start with you. W- What's important about that?
>> So, I'll start with just generally uh because it again, as a buffetologist, he always says, "Well, pretend that you're buying the entire company. How much would you expect to get out of it?" And that's what you would use to value the business. Ultimately, it's just the free cash the present value of discounted free cash flows, right? Okay, fine. For the hyperscalers to me, I got to put an asterisk on this cuz it's not only that, the example we were talking before with the three of us bought meta, but also when we think about the level of CapEx that they're spending, they're saying, well, look, we generate all this cash flow. We can now invest in all this CapEx. No problem. And I'm sitting there thinking, no, but it is a problem because you're telling me that you're making $45 billion when you're really making, you know, as me and John would say, like two or three. Right? Two or three billion. That's what's actually left over for the owner, but your CapEx is going up, let's say, 30, 40, 50% a year. So, where do you make up the difference? You start using your balance sheet. So, all of these companies, Microsoft, Meta, Alphabet, you name it. Now, they're taking on a lot of debt. You can make the argument that if they didn't have to pay cash stock compensation expense, you wouldn't have had to borrow. You wouldn't have to lever your balance sheet. Alphabet is also unique cuz as you know, they did 30 billion in borrowing in Q1 of this year, and then they just issued 85 billion in shares. And so, the fact that they're using their cash to pay employees with stock-based comp means that they now have to go use their balance sheet to a degree they wouldn't have had to otherwise, technically. And so, I have no problem with that. Run your business the way you want to, but present it a little more cleanly, so people understand, well, okay, part of this is not just AI data center build-out or whatever it is. Part of this is compensation. And as an investor looking to value the business, I need to know what the real number is that I should use.
>> As you're speaking, Kevin, I'm sitting here thinking, okay, so they're using debt. So, what? I use debt, you know, I have a mortgage, I have a credit card. Why can't these giant companies with, you know, incredible legacy businesses, right? Why as an investor should I be that worried that these companies are are going to need to use more debt in the future.
>> One, you might not be able to see all of it because a lot of it's being done off balance sheet. Uh and then directionally, if you think of Meta uh just as an example, we keep on picking on them, but 5 years before they had no debt. These were all asset except with the exception of Amazon. All of these hyperscalers were asset-light companies uh until very recently. Now, what that meant is, you know, they're they're not issuing a lot of debt, if any. Uh they don't have a lot of hard assets, you know, physical plant, property, equipment, that type of thing on their books. Um
>> Before they didn't.
>> Before they didn't. Now they're Now they're they're they're racing. It's a almost a land rush to, you know, to build out the infrastructure they need to um have this technological capability. What they're going to do with it in the end, nobody really knows yet. But it's one thing when um their legacy businesses are generating enough cash that they can finance the build-out. It is another thing when you when they realize that we actually our legacy businesses aren't generating enough cash to finance this build-out and the capital needs to to build it out are growing and growing and we really don't know where it's going to stop yet. Now you're getting the point where they're starting to hit a wall and they need to get more creative in how they raise money and then having to cut out cut expenses, lay people off. Um and it even start tolerating dilution, not doing as much in the way of buybacks. That's a big change culturally as far as the structure of the business model and I think we're going to start seeing more companies uh probably doing stock issuances to issuing more debt and until this land rush phase is over with and I don't think anybody really knows how long uh it's going to be sustainable until somebody ends up having to tap out and say we can't keep up anymore.
>> All right, so with what you've just outlined here, do you think the market has taken that into account? Do you think that the forward PE ratios reflect the nature of these businesses as you guys have been describing?
>> Not even a little bit, I don't think. And and the reason I say that is because when you when you think about how many companies, let's say, are valued off of the EBITDA. EBITDA doesn't account for this. EBITDA is just as it this the cash cost of stock-based comp isn't in there. Most companies report adjusted EBITDA and they add back stock-based compensation. So, if you want to use enterprise value to EBITDA, it's not in there.
>> And EBITDA, of course, is a is a common earnings measure that Wall Street uses and that stands for earnings before interest, taxes, depreciation, and amortization.
>> Correct. Here's the thing I worry about. When you think about it even from, let's say, a credit risk perspective, most of the time your debt covenants are based off of net leverage ratio, your numerator, right? Or sorry, your net leverage numerator. Uh net debt divided by EBITDA. Well, if the EBITDA doesn't account for this expense, then your net leverage ratio is understated. So, that could impact your credit rating, which could impact your interest rate. And then, as Buffett always says, you know, it's not necessarily intrinsic value, it's intrinsic value per share. And the dilution that John is rightly pointing out, we might be growing the size of the business, but the the stockholder is not benefiting at all because we just keep diluting them. And so, you have to be careful with this.
>> Well, a couple points, too. Uh if you look at the sell-side analysts, the Wall Street analysts and their projections, there's this miraculous, almost checkmark-shaped uh recovery to free cash flow that's being projected right now. So, that you're going to have basically a collapse to very little uh free cash flow among the major hyperscalers this year and next year. And then a sudden rebound to 2025 levels. And then almost a near doubling after that uh in 2029. Uh so, that it's far surpassing the free cash flow. Now, the notion
>> That's like the analyst consensus right
>> Yeah, and and the notion being and it just seems improbable, especially once you start getting to out year, you know, estimates. It seems improbable that all of this spending is going to uh be done with you know, at the end of 2028.
>> So, you don't think it's likely that we go through a spending phase, right? Everybody puts shovels in the ground, they build their data centers, they hire their engineers, they build up a a model that the data is sufficient to kind of sell to everybody in the world, and then after that they begin to reap some of the benefits, right? Like you you you get the benefits of all that previous spending, you get some improved unit economics, right? Like what that that doesn't seem that implausible to me.
>> Well, the CapEx would continue to spend would continue to grow.
>> not at the same level.
>> The tapering off. That's what the But nobody knows. Nobody Nobody knows how this is going to play out. So, if you're looking at multiples, my guess is that people are probably looking out to 29 2029 2030 numbers. But again, it's very hard to predict. You know, 4 years ago nobody would have predicted it was going to turn out the way it did. I don't know how anybody predicts now what it's going to turn out to look like in 4 years. But if you're talking about, you know, where are these where is the disagreement on, you know, how to value these things? It's probably going to be on the out year numbers.
>> And this this is actually if I can piggyback off you, John, this is actually a really interesting point because I I believe it was Colette Kress, the CFO of Nvidia. She recently said that look, by 2030 we believe that required infrastructure spend for data centers will be 3 to 4 trillion per year. And I looked at that and thought, well, to John's point, if the analysts are forecasting a rebound in free cash flow, think of the amount of cash from operating activities or operating cash flow that would need to pick up to offset that. Cuz we're not spending anywhere near that level now. And I thought it was a funny number to say 3 to 4 trillion because global IT spend right now, at least thanks to Google, is about 6 trillion. So, they're saying that global IT spend would have to go up by at least 50%. So, to John's point, when you think of the magnitude of operating cash flow that would have to come up to offset that, it it doesn't seem like the sell-side community is taking that three to four trillion in either, and it came from one of the Mag 7 CFOs.
>> You're also seeing now uh a lag time between the time that let's just use Nvidia as an example. They're selling chips to all the AI companies. You have the you know, semiconductor makers and other makers of AI hardware, Nvidia, and the like. Um their earnings and revenue are benefiting immediately, but their customers are able to defer this expense over a number of years, and in some cases, they're not even having to start the depreciation expense because it's going into data centers that are under construction, and the depreciation period doesn't even start tolling until they turn the lights of the data center. So, in the meantime, you're going you've got this period, and nobody knows if this this golden moment's going to be 12 months, 18 months, however long it goes. But, yeah, where the entire S&P 500 earnings are getting lifted because the benefits of the sales are coming through, and the expenses are not.
>> All right, we're going to take another quick break, and when we come back, more with Kevin Coharki and John Weil. So, just to wrap up, you're talking about something pretty profound, really really calling into question whether or not the earnings that we are seeing out of all these companies, which look enormous. People have been talking about how we're in this incredible cycle for earnings growth. And John, it makes me think of something you wrote in your piece about Meta when you were writing about they look like this money-printing machine, but if you look at the free cash flow more closely, uh it becomes quote you said something of an optical illusion. Uh thinking about that phrase, Kevin, how widespread would you say people should be concerned about the earnings that they are seeing today being not really what they're what they look like on paper.
>> I think I I very concerned, and uh with a lot of the consulting work I do, I get to work with a lot of engineers that are involved in building some of these things or using this technology. I was on the I was on the phone yesterday with the chief technology officer of a $90 billion a year company and we were talking about this exact thing. We were kind of just trading opinions back and forth and I said, you know, one of my problems, sir, I'm not going to say his name obviously, was a sir. I said is, you know, you always want me to train your engineers about life cycle cost, right? What is the cost from creation until death? And I don't know that anyone has factored this in yet. What I mean by that is these data centers are going to suck up so much water and electricity and everything else. There will be an ultimate cost to that later on if there are environmental impacts. We've seen this with tobacco, asbestos, all this other fun stuff. How do you
>> Costs that we haven't even conceived of necessarily yet, yeah.
>> Correct. Correct. And in government contracting, this is a very key thing. You have to account for the whole life cycle cost of whatever it is you're building. And so how do you price that into your product? One, I don't think that it has been. And if it's not priced into the product, then it has not been priced into the free cash flow that investors and analysts are looking at. So I don't I don't think there's a full understanding of this at all. Coupled with the fact that a lot of my engineers are saying, look, we normally build a building for 30 to 40 years, but the energy usage is so strong, a lot of these buildings maybe they'll last 15 years. And I just think then that changes the CAPEX cycle considerably because it obviously quickens it, which means the reinvestment needs are just higher. And I just don't think that this is very well understood yet by market participants.
>> So, Kevin, I know you're not here to give investment advice. But but but if you're you know, if you were talking to a friend and they were saying, you know, I'm worried about this AI bubble, would you would you say that their bearishness is warranted? Would you advise people to to change their investing strategy to sort of see if they can minimize the impact of this stuff in their portfolio?
>> I would advise them to at least reconsider maybe some of their prior thoughts. So, funny enough, I am a registered investment advisor. I have a series 65.
>> There you go.
>> So, I could say it legally. But, um I wouldn't tell anybody specific advice. I wouldn't give them uh stock recommendations. Obviously, that's not what I do. But, I would tell them on the analysis just think about it a little bit more deeply and not just take the reported numbers at face value. Which is one of which is the main reason, honestly, I called John because I've been reading his If I can give him a hat tip, I've been reading his articles for years and thought if there's anyone who's going to understand this, it'll be him.
>> The other side is that it's a bull market. Bull market, you put multiples on hopes and dreams. And when things can't be quantified and everything's a debating point, you're going to have optimists who look at this and say this is a breakthrough technology. We don't We've seen what breakthrough technologies can lead to in terms of wealth creation, and we don't know where this is going to go, but we're willing to to bet that the people who are building these things are building it for a reason. And sure, they may find out later, you know, they may be waiting to figure out later exactly how they're going to monetize this cuz they don't know yet. But, we'll take the chances on them. And that's, I think, as much as anything what drives up stock prices and why these companies are out able to go out and borrow at very low cost of capital. And it will work until it doesn't.
>> Yeah, the market certainly is has has been there to to finance this so far. So, um a lot of people think that that maybe they can all collectively land the AI ship. All right. Uh both of you, thank you so much for joining us. Kevin, thanks so much.
>> No, thank you very much for having me. It's been my pleasure to be here.
>> for having me. It's always
>> And John, thanks for thanks for coming down from the sixth floor here to the fourth floor with us.
>> Thank you.
>> [music] [music] [music]
>> All right. So, first of all, I got to say congratulations cuz it's only a handful of people that have pronounced my last name right. So, wait a minute. Well done.
>> [laughter]
>> I didn't You know what?
>> What did I
>> Shame on me for not doing
>> No, you said it earlier. You said it earlier. I heard it. I was like, "He's got it."
>> [music]