The Dividend Cafe - The Real AI Problem Made Simple

Episode Date: September 4, 2026

Today's Post - https://bahnsen.co/4hbt99a This special edition of The Dividend Cafe argues that the key issue investors are missing in the AI story is not AI’s usefulness but whether massive AI infrastructure spending will earn an adequate return on invested capital. The episode highlights how major tech firms are raising unprecedented debt and equity—despite strong earnings—because free cash flow is falling or turning negative under enormous AI compute and data center CapEx. It notes high customer concentration and interconnectivity across the ecosystem, including Nvidia’s revenue reliance on three customers and AI labs’ heavy dependence on a small share of customers, alongside purchase commitments far exceeding current revenues. The central risk, the host argues, is whether capital markets continue funding the buildout and on what terms before profitable utilization arrives. 00:00 Welcome to Dividend Cafe 00:58 AI Everywhere Now 04:02 The Missing Investor Issue 06:16 Capex Funding Frenzy 07:59 Earnings Up Cashflow Down 09:44 Off Balance Sheet Reality 12:14 Why AI Economics Flip SaaS 16:28 Unknowns Behind Monetization 19:11 Return on Capital Question 20:58 Capex Bubble Spillover Risk 22:50 Concentration and Connectivity 24:10 Capital Markets Are The Gate 28:32 Wrap Up and Disclosures Links mentioned in this episode: DividendCafe.com TheBahnsenGroup.com

Welcome to the Dividing Cafe weekly market commentary focused on dividends in your portfolio and dividends in your understanding of economic life. Hello and welcome to this very special edition of the Dividendin Cafe. I say it is special because it is a topic that I have believed for some time is one of the most important topics in all of investing right now. I am particularly pleased with this week's Dividing Cafe. I worked very hard on it to reduce all of the things that I could have been distracted by into the singular topic of what the one major issue is. All of this AI talk that I think investors are missing, that I think investors need to understand, and that we are going to laser focus on today in the Dividend Cafe.

Let's start with a basic caveat. I think that anyone paying attention to the just the day-by-day noise of markets, the sort of investor ramifications of the AI story, if you read the Wall Street Journal, if you watch financial television, it's rather abundantly clear that AI is this huge story at various levels of engagement, nuance, particulars, but you really can't be paying attention to investment markets without hearing AI as more or less. the primary story. Some of them are more interesting than others. Some are more intelligently presented, but there's just an abundance of coverage about AI for very good reason for anything related to investors and markets and all that. But then this is also very true of any discussion of the economy. When we look at the state of jobs, when we look at the state of jobs, when we look at the state, of GDP growth and the sources of GDP growth, when we look to understand potential ramifications around business investment, capital expenditures. These things are all

just unbelievably intertwined with the AI story and discussion. AI is not only the kind of major investment story, it is the major economic story right now as well. well, even more so than traditional economic story favorites like the Fed and interest rates and the midterm elections and things like that. And then speaking of those midterm elections, this third category I want to mention is in the political sphere. The AI story is everywhere you look. All of a sudden, lo and behold, there is bipartisan populist rage over data center construction.

There is various left-wing critique of the AI movement because of the billionaire tech tycoons and oligarchy that surrounds it. There is right-wing opposition because of the lack of trustworthiness of the big tech folks or privacy concerns or whatnot. It is a reasonably unavoidable story in the political sphere as well. And so whether you're talking about investment, economic, or political domain, the AI thing is touching all the above. And all of these different categories in our society where there is AI discussion, create their own levels of noise and their own levels of heat that is often very different from light. And that's fine. I'm not here today to say, I'm going to clean it all up. but I do want to talk about the important part to us in the Dividing Cafe. To the extent the Bonson Group has an investment point of view,

and there is an element to this story that I think is being missed. Now, there's a lot of elements to the story. In the written Dividendinacethafe.com today, I actually went through and I think found the last 10 dividend cafes I had written about AI in one form or another, and provided the link where we're talking about AI and jobs, where we're talking about a bubble risk in AI, where we're talking about inflationary or deflationary

ramifications of AI. We're talking about the history of various bubbles, when we're talking about what disruption AI is creating and is not creating in the software space and so forth. So we've done a lot around AI here in the Dividy Cafe, and that's because it is such a big story for investors in the economy, in the political sphere, et cetera. But what I want to suggest today is that there is a better way to narrow down the big issue right now than what has been done. And I

include myself in this, by the way. I wrote my 2026 year ahead paper, and I think we had nine forecaster themes in there. And number one was 2026 being a year in which AI vulnerability were going to be a thing. And quite specifically, AI vulnerabilities will become much more evident to markets. That was the number one theme in my 2026 white paper. And yet, I focused in that on valuation excess, on a low regard for risk, on the inability for every one of the companies competing with each other to all win at the same time. I focused on China risk.

I focused on cultural political pushback. But the thing I'm going to kind of more specifically talk about today was maybe tangentially mentioned. I have a half a sentence that I could extract that you could say is in this domain, but we're going to do the right kind of dive into it today. So let me start by telling you that in the last several months, you have seen in American financial markets a $500 billion debt facility be announced from

NVIDIA in conjunction with the consortium of Wall Street asset managers and lenders, a $85 billion common stock offering from the company formerly known as Google, 100-year bond offering from the company formerly known as Google, Alphabet, as we know it. Intel, doing a $15 billion common stock offering, specifically to fund additional CAP-X needs. Oracle announcing a $20 billion equity raise and a $25 billion senior debt offering. Amazon announcing a six-part bond deal, raising $25 billion in a new. debt, META, the company formerly known as Facebook, announcing a $30 billion bond deal, their first bond offering in years.

And I am skipping in this list. By the way, everything I just announced, there's a link to those deals in divinitycafe.com, and I am purposely skipping any of the other debt and equity capital announcements that were specific to a particular data center. Okay, that would make the list even longer. These are just general debt and equity raises related to broader AI CAP-X at a company level, at an enterprise level. And I want to start with a question today. All of these stocks are up because they are all making more money than ever.

That's what we understand, right? Earnings are huge. Okay, fair enough. But if they're making more money than ever, why do they need to raise more money than ever? That's my question. And the known answer, the highly public answer, is that their free cash flow is actually collapsing, in some cases going negative. Revenues are increasing, but cash flow is declining. It's that simple.

And the reason for good or for bad, right now I'm just doing a basic accounting answer, the reason is because it is expensive to make this stuff. this AI stuff, the infrastructure, the build-out that is necessary for not only the existence, but then the subsequent applications embedded in AI. It's very high. These companies do not need capital for general corporate use, but the expensive cost of AI compute, that is the thing. And what we really see more and more is the amount of money needed, for AI infrastructure buildout is funding the very customers doing the building.

Okay? So it touches that circularity thing we've talked about in the past, but the point being we're living in a time in which massive amounts of burden are being put on capital markets to fund something at a time when revenues are already very high, but cash flow is declining and this is going to lead to our subject today. Now, one thing I want to point out is I listed a bunch of debt and equity offerings that do not tackle so-called off-balance sheet issues, and I avoided it on purpose. I'm not trying to make it look better than it is, but I am trying to avoid making it look worse than it is. There are plenty of folks that might want to sound alarms on this whole subject that immediately go to the fact that you're probably at about a trillion and a half dollars of off-balance sheet obligations as well. you go, well, David, that seems relevant. Why are you skipping it? And it is because I do believe

many that are sounding that alarm are doing it disingenuously. I think what I'm going to bring up today has so much concern in it that I don't need to melodramatize it or be disingenuous with somewhat misleading statements. The trillion and a half number skews the fact. First of all, let's point out that stuff on balance sheet is concerning enough. Okay, we don't need to go pile on. But just secondly, it's not really the same as a current liability. It sits on a balance sheet. That when there's a cash outflow commitment into the future connected to future revenue

expectations, there is a different discounting of that for good reason. I think many who repeat the off balance sheet numbers are doing it for shock and awe purposes. and I will challenge people that have followed Dividing Cafe for any period of time to say that I do shock and awe here in Dividend Cafe. I really believe I don't. And if I'm fooling myself, then forgive me. But I think that's true. Now, the other piece is that these off-balance sheet obligations in many cases are cash outflow commitments that go 15 and sometimes 20 years. If you were to do a net present value of those, it would be a much smaller number. and take away a lot of the melodrama people are after.

So I don't want to deny there are substantial off-balance sheet obligations, both lease commitments and as well as GPU supply deals. I think the data center lease commitments are probably the biggest. But I think that we can avoid exaggerating it and just speak to stuff that's a little cleaner, and that's what I'm doing. Now, from a peer business model standpoint, Nobody disputes. You can be the biggest AI bowl or AI bear. Everyone understands that the large language models are expensive to run. And that is kind of at the heart of the matter here. Their business models are essentially very similar to one another. There are nuances. There are a few product differences, but their underlying funding is more or less very similar. And I think what's noteworthy is how. totally turned on its head, the LOM story, these AI labs, from what we have become very

custom do in thinking about the whole technology investment story in America. The whole software as a service story, where essentially we've been very used to companies that as their revenues scale up, their costs don't move. And in this particular case, as demand for AI is going higher, the cost are increasing even more so. It is very upside down relative to what we've thought of is the appealing economics of technology investment in the last couple of decades. Look, I could point out the fact that like a lot of dot-com warnings in the 90s, you said, oh, these companies have no revenues, no earnings or inadequate revenues and earnings and their valuations are way out of skew. But this is a little bit of

bit different in that while we're referring to here at the AI Labs, I'm not critical of the fact that they generate a certain amount of revenue now and are spending astronomically more than that. I'm not really talking about the present spend and the present revenue. I am saying maybe we have a question about the future spend and the future revenue. And I will take for granted that they have some way to make money into the future on these things. But I am unaware of what the plan is for increasing revenue and creating the scale that we're used to. You know, the fact of the matter is that this is a very expensive need to get that revenue and there

is no sign of that expense coming down into the future. The AI infrastructure buildout is hundreds upon hundreds upon hundreds of billions of dollars that is going just to hyperscalor CAPEX need with data centers with GPUs with power, with cloud, cloud capacity, networking, equipment, and so forth. And it is nowhere near being covered by the actual AI generative revenue or future pro forma expectations for it. Now, I don't think anything there is controversial. The question that I believe is reasonable is what in the future revenue will rationalize it,

where will it come from, why will it come? And I want to suggest some potential answers. Some AI critics will attempt to answer that question, but with an AI skeptical answer, it is never going to come. The public will never allow it. The companies have overpromised. They're going to under-deliver. etc. I'm not saying that. And what I mean by that is I'm not necessarily saying that. I'm not not saying it, but what I am limiting my scope here is to the incremental profits of whatever

AI does end up doing. And I think this is a very important distinction. I do not believe the current capabilities or even the potential promise of AI's business use, adoption, application, are really the major concern. I acknowledge that there's additional risk in these things. I think that there is going to be worse financial ramifications if they do under-deliver. But I'm willing to bet that doesn't happen. The issue is, are we sure the features and capabilities

that will come with the needed revenue growth and incremental profit growth will justify these levels of investment? Okay. So I'm going to dumb this. down for you here in a moment. To attempt to answer this question, we have to know a few things that we just don't currently know, okay? The questions cannot be answered without being able to say what the actual customer cost will be, the economically sustainable cost. What does it cost to have a customer in a sustainable way where the customer is paying their full freight,

the provider's not losing money on that customer? So we're not talking about a subsidized customer. What is that cost going to be? That number is a complete mystery in the AI lab business story right now. Number two, what the future utilization will be, how many customers at what level of usage, what level of monetization is going to materialize. Again, it's an unknown. People can put in different projections and some optimistic ones may prove to be accurate. But I would like to point out that there are many who are far more bearish than I am using negative inputs here, and I cannot tell you that they are wrong. But on the flip side, whatever your inclination, suspicion is about the future of this, it is unknown that I think is striking how wide

the dispersion of potential results is. You cannot, by the way, say, well, everyone is spending so much on this, that's the answer as to how this thing's all going to play out. You cannot say it makes so much sense that everyone is spending all this money because everyone is spending all this money. And I do believe that is the most common answer out there, even if it gets worded a little differently to sound better than that. I'm not used to hearing that everything is going to be all right because they're all doing it as a bull case. That strikes me as wishful thinking in a lot of ways. But the third issue that I think needs to be understood is if the cost of having a customer

comes in at an acceptable place and the answer to what AI ends up delivering in terms of utilization is better than expected, will those two things happen in a timeline that is palatable, the assets involved to feed these things are depreciating assets, right? Will the utilization that is profitable? Will the profitable utilization we're trying to get to happen before the technology is obsolete? And embedded in these unknowns get you to the heart of the matter

the subject of today's Dividend Cafe. And I'm going to give it to you, and I think this is eight words, what will the return on invested capital be? This is the question. What will the return on invested capital be? We don't need to question the usefulness of AI. I don't. We don't need to question its transformative wonder.

I don't. But we have to question whether the usefulness and the transformative generation out of AI, will I'll generate the return on invested capital. And anyone who tells you that question doesn't matter is not offering an investment opinion. They are asking you to join a cult. It is at that point become something almost more faith-based than anything in the realm of rationality. Now, I understand that many would say, you're doing a present tense observation and you're not really understanding how revenues and cost in the future are going to be wonderful.

And you don't know that cost won't come down as users and user death increases. The technology is going to so outperform expectations that we just need to live a little. Assume the best, the monetization will outperform. And by the way, the technology thus far is outperformed expectation. why not assume the monetization in the future will outperform? Again, I guess that's possible. I don't know that it's an investment argument as much as a religious one, but okay. But here's the thing I would say.

The massive overspend that some are concerned about right now, if it all proves to be a major overspend, you cannot justify the amount. moment by saying, hey, I'm invested in the companies receiving the overspend. It's okay. This is the sort of hyperscaler versus pick and shovel argument that the spend, when we talk about it with one side, is revenue on the other. And of course, vice versa. The revenue of let's say Nvidia is the spend of some of these other companies, whether they be AI labs or hypers. But remember that even in the stock prices of those receiving the money in this buildout,

there is expectations of sustainability and growth of that spend that if it were to not materialize because the whole spend ran into question, then it would end up not merely hurting those doing the spending, but those who are counting on that spending going forward. This, of course, is the story of the tech bust and many other busts beyond the tech and telecom bust in the last 25 years, but most CAP-X bubbles in the last several hundred years. It was a cascading effect that hit all comers, the receivers and the givers. And so I think where we're headed with this is a question about what could catalyze some concern, some problem, some reckoning of this spending

and potential excess in the spend. And that brings me to the financial markets story. Nvidia's 10Q, by the way, I have a link to it in Diving Cafe. Flat out discloses. Three customers are 54% of revenue, 21, 17, and 16 respectively. Three customers, the same three,

are 30, 18, and 16. 16%, 64% of their receivables. That's not a very diversified customer base, but what about the diversification of their customers? And here I'm going to put a chart up on the screen. When you look at Open AI and Anthropic, that 80% of their revenue comes from 1% of customers, 1% of businesses.

There is, in the customers of the lab, all the way to the customers, the labs themselves of the pick and shovel companies, there is a massive concentration and interconnectivity that is exponentially riskier than anything we've ever seen, ever. And these major AI labs have purchase orders, this capital spending commitment, have purchase orders into the years out into the future over 10 times. total revenue. So what we have is just indisputably a question about capital markets. Because I started off today's discussion saying capital markets are now having to come fund this whole thing, major debt issuance, major equity issuance, financial markets coming in to

feed this. And I guess what I would say to you is that investment in the AI ecosystem right now has almost nothing to do with how AI as a technology does, how it's utilized, how it's monetized. It's a financial story on whether or not capital markets allow it to play out. Because if financial markets pause, if they adjust their expectation for terms, if there is some sort of a hiccup around qualification, around structuring,

the receivables are not there to pay for it. And this is always and forever the story. Financial markets funding something that can't pay for itself works until it doesn't. Now, again, people may say, no, it is all going to work. It will hold up. And I am not saying it will not, but I would say it would be the first time in history if it were to happen without any pauses, hiccups, disturbances along the way. Look, the receivables will not matter to those holding the bag.

in the current economic story of AI. So if the lending costs go up, if what is needed to clear, you know, basically the cost of capital, equity investors are presupposing, if those equity terms get re-rated in the market, the domino effect would be utterly extraordinary. And I don't think that requires order flow to slow down. This is one of the great disingenuous things is every time a new quarterly result comes out and they go, wow, it was even bigger quarter, Amazon spending even more than we thought. Wow, it's even bigger quarter. Broadcoms receiving more money than we thought. Those things are not news stories. They're completely expected. The question is how it's

being paid for. The answers is it's been paid for from capital markets. What if capital markets start to call into question the future return on invested capital? The ability to generate incremental profits in the future from all of this spending. The fall, the fall, A far larger risk than a slowing of order for this compute power is A, the access to the money to pay for it goes away, or B, the access to the money to pay for it changes, as far as the terms involved, or C, and by it could be and or, the profits to be derived from these expenses, the revenues to the pick and shovel companies that are expenses to those paying them don't materialize. I am not saying A will happen or B will happen or C will happen, but of that A, B, and C, I will tell you that all three are possible. All three could end up being problematic, and it only takes one of the three. I would say historically, a common sense analysis, I expect that there will be major problems there. I am too humble to tell you what, when, where, how exactly.

But I think it's a pretty fair point to point out that this is underappreciated in the overall story. And I want to say this too. I'm avoiding the various other subjects about data center popularity, about the kind of political environment around it. I'm avoiding which AI labs are going to beat out other AI labs in a competitive landscape. There is a lot of other things that get talked about a lot and that should be talked about that are not really my greatest subject here. I'm saying to you that the core vulnerability is whether or not these profits are, going to rationalize the largest cap-x boom in history and whether they're going to come in time. And when I say in time, it's not a completely subjective phrase because the timetable is set by the only timekeeper that's ever mattered in business history.

I'm talking about the timetable set by markets. Thank you very much for listening, watching, and reading Dividing Cafe. I really hope you understand the fundamental point we're making here today. I welcome any and all questions as we get ready to go into this three-day weekend. Enjoy your weekend. Enjoy this bridge out of what has been a very hot and human summer into what we hope will be a wonderful fall. And please do always remember you can send your questions to questions at thebonsongroup.com. so much for being a part of Dividend Cafe.

The Bonson Group is a group of investment professionals registered with Hightower Securities LLC, member FINRA and SIPC, and with Hightower Advisors, LLC, a registered investment advisor with the SEC. Securities are offered through Hightower Securities LLC. Advisory services are offered through Hightower Advisors, LLC. This is not an offer to buy ourselves securities. No investment process is free risk. There is no guarantee that the investment process or investment opportunities referenced to Rian will be profitable.

Past performance is not indicative of current or future performance and is not a guarantee. The investment opportunities referenced herein may not be suitable for all investors. All data and information referenced herein are from sources believed to be reliable. Any opinions, news, research, analyses, prices, or other information contained in this research is provided as general market commentary and does not constitute investment advice. The Bonsor Group in Hightower shall not in any way be liable for claims and make no express or implied representations or warranties as to the accuracy or completeness of the data and other information, or for statements or errors contained in or omissions from the obtained data and information referenced.

The data and information are provided as of the date reference, such data and information are subject to change without notice. This document was created for informational purposes only, the opinions expressed, are solely those of the Bonson Group and do not represent those of Hightower Advisors or any of its affiliates. Hightower advisors do not provide tax or legal advice. this material was not intended or written to be used or presented to any entity as tax advice or tax information. Tax laws vary based on the client's individual circumstances and can change at any time without notice. Clients are urged to consult their tax or legal advisor for any related questions.