AI's Bar Mitzvah Moment? From Hype & Hope to Business Questions!

Hi, welcome back. I'm old enough to remember big changes coming to business and personal lives. PCs in the 1980s, the dot in the '90s, social media 20 years ago, but I've never I I don't remember any of those ideas being as pervasive as AI has. It's crept into every business conversation, every investing conversation, and many personal conversations. Now, if you have a selective memory of history, you think of AI as being a recent phenomenon. In fact, many people think AI was born November 30th of 2022, which is when Chat GPT was unveiled to the public. We'll talk about why that was an eventful, you know, it was eventful for AI. But I think we need to step back. AI has been around a lot longer. It's been building as long as we've had computers. In fact, uh I remember when IBM came up with Deep Blue. It was a computer that they fed every chess game ever played into. It

was able to look at 200 million different chess positions per second, and it beat every grandmaster it played. On the cultural front, we've seen books and movies centered around machines that are not just powerful, but so powerful that they take over the human masters. My favorite is 2001 of space odyssey where Hal tells u you know Dave his controller I'm afraid I can't do that Dave and you get sense that that conversation is playing out with a lot of people as they talk to Chachi PT or Claude or with Grock now if to see how much of an explosive effect AI has had I mean let's step back in the last four years the most successful company in terms of adding market cap in the world has been Nvidia, a company that provides the chips that make the AI revolution possible.

Until about 2021, I was talking about the fang, Facebook, Amazon, Netflix, Google, Apple, and Microsoft. But with AI taking off, it's become the Mag 7 where Tesla and Nvidia have replaced Netflix. And those Mag 7 stocks have carried the market. They accounted for about 45% increase in market cap of all US stocks between 2022 and 2025. And their aggregate market cap just those seven companies was 23.7 trillion. Now it's not just markets that have become disordered with AI. Companies have invested massive amounts into AI. Many of the mag seven companies that I talked about are companies that have put in know hundreds of billions of dollars even trillions of dollars in AI capex and it's estimated that that capex has sustained the US economy in fact between 08 to 1% of real growth the real growth in the US economy is only about 2 and a.5% has come from AI investment AI is

clearly here to stay now if you look at the AI story it goes beyond business and markets My wife is a fifth grade age teacher and I remember her coming back a week after or a few weeks after chat GPT was unveiled talking about how chat GPT had entered the classroom and as people look at AI tools and um and agents there are two reactions you get. One is awe at what these agents can do and at the same time dread that these agents can take over their lives make them obsolete.

Now, it might just be me, but I have a feeling that the AI story has gone a little bit off the tracks in terms of public perception. When it started in 2022, maybe even 2023, spilling into 2024, it's mostly an upbeat, positive story, a tool that was going to make us all productive and more efficient. Sometime in the last year or two, at least, and it might be my imagination, it's taken a negative turn. It showed up for me when in graduations across the US in 2026 when speakers at graduation ceremonies were booed. You know how difficult that is to get done because at graduation ceremonies people are happy.

It's tough to make them unhappy enough to boo you. What these speakers shared in common is where they're talking about how AI could replace jobs. Now as this has all all unfolded there's been a massive discussion of AI's potential and promise in business but in spite of all the people talking I have a sense that the debate has gone again is is gone off the tracks because people seem to be talking after uh past each other. What am I talking about? AI optimists keep talking about the huge potential market for AI products and services. AI skeptics keep talking about how many hundreds of billions of dollars of we invest in capex and how that investment is too large to be made whole again. Now there's an element of truth to both of these groups but I think neither group is telling the whole story is willing to look at the whole story.

The AI optimist may be right about the market being big but big markets don't necessarily translate to valuable businesses and valuable companies. And the skeptics might be right about the hundreds of billions of dollars of capex as being a big number, but that by itself doesn't create a bad business. What I would like to do in this session is talk about closing that loop, bringing together the two sides so we can actually have a discussion that is centered in reality. Now, as I as you listen to me over the next 30 to 35 minutes, I have to make a confession. I am a very light user of AI. I have the free chat GPT version and I use it maybe once a month, twice a month. I don't use it in my writing. I And that might be because I'm old and a little eccentric.

No, I've never I'm just a passing acquaintance with Claude, though I know that Claude agents, especially the banking business and the valuation business, have drawn on my books. I'm writing this book for an write I'm doing this session for an audience of one myself. And here's why. As I listen to all these different debates and as you'll see there on different tracks, I'm getting torn in a dozen different directions. I need a framework to make sense of all of these different sides. And that's what I'm aspiring to do in this session. So let's start the discussion by talking about what happens when there's revolutionary change in business or in you know or in investing.

Now I you know I I I may be over overgeneralizing here but almost every revolutionary change goes through four phases. There's the hope and hype phase and during the phase you have visionaries selling hope that this change is going to be big and they do it very little tangible to point to. At the same time they're evangelists. They're selling other people that the change will come. And why do they need to do that? to get people to work for them for the change and people to supply them with capital. But if you look at the actual entities they're creating, there's nothing much to look at there. There are no assets.

There's definitely no revenues yet. Lots of expenses. That's a hope and hype cycle. Then comes the investing buildup. The change has become entrenched enough that people are willing to put money in it there. This is where you get startups and young business. you're investing in capacity for a change that might or might not come. The risk here is you can invest in the wrong architecture. In fact, I've talked about the big market delusion where faced with big change, people often overinvest because they're overconfident people and they think the change can deliver more than it can. But you get a capital buildup. And if you just looked at these companies, they look like basket cases. They've invested billions maybe tens of billion dollar billions of dollars in capacity but they don't have much to show in terms of revenues and profits. Then comes the phase where you build businesses. This is where you start to generate revenues and earnings. You try different there's a lot of trial and error because you got to figure out what works and your

revenues start to grow if you're successful and your earnings start to show up a little later but you start to separate the winners from the losers. By now the change is is in the process and you start the reccalibration where the industry kind of resets itself, decides where winners and losers are going and sees how this will play out. Businesses restructure and it's also where you see the dark side of disruption. What is that? Change always creates damage to the status quo. Those companies leave, you get a new status quo with the change. We're built around the change and companies look for modes and competitive advantages. Revenues and earnings start to level off because the business is now mature. You see, we're done. Not quite because the cycle starts all over again because you're now the status quo and change comes after you.

Now, where is AI in this change? The remember I talked about chat GPT getting the process rolling. It had a key role to play in AI becoming what it is because it made AI accessible to everyone. So while it's low tech AI, it started the hype cycle rolling. Over the last four years, you've seen the hope cycle build up and then the investment cycle build. In fact, we've never seen a buildup as large as the buildup we've seen in AI in terms of companies building the AI architecture. Now the companies that supply what you need to build the architecture have benefited.

Nvidia with chips know LLMs that provide the the the software to create these chips and power and data into products the power and water companies and but for for now we're just building the the architecture and in fact this is where the hyperscalers have entered in in terms of how much money they put in the air cycle. Remember I talked about the mag 7. If you take four of those mag seven companies, Meta, Amazon, Alphabet and Microsoft, NATA, Oracle and Core Wee to the M six companies out of the thousands of companies list in the US just these six companies. If you look at how much they've invested between 2020 and 2026 have invested close to 1.73 trillion in capex, much of it in AI. And in fact, if you bring in the capex from other companies, smaller, less lower profile companies, we're going to get to two trillion, two and a half trillion. The way I describe AI

as it stands now is we built the largest factory in history, most expensive one, more expensive than the railroads, the automobiles, the and we've done it in hypers speed, but we don't quite know what those factories are going to produce or whether anybody will buy them. It's a dangerous place to be. But let's think about the business building that is coming to AI. And I think that's why I call this the bar mitzvah stage the process. Bar mitzvah of course is the Jewish right of passage where you take essentially teenagers and you tell them that they're grown up with all the pluses and minuses of being a grown-up.

They're not quite ready, but you push them in there. I think we're reaching the bar mitzvah stage of AI where you can't justify things based on it's going to be a big market. We have to start talking about business building. Now, generically, if you look at business building in any business, there are three prongs that drive the story of a business. The first is you need revenues. Those revenues come from having a large market, total market size, and the share of that market that your company can get at that market side. It comes from profitability because revenues themselves don't create value. You need to make money. So in terms of what you generate as margins and their unit economics come into play.

What are unit economics? It's what you make on the next unit that you sell. What does it cost you to produce that unit in economies of scale? And then thirdly, the reinvestment plays a key part in the story because you can generate a lot of profits but if you require insane amounts of reinvestment might not be a good business. So there you got to look at the capital intensity of the business you have and how much lag there is between when you invest and when you can get revenues out. That's how any business is bu built. Let's take these three legs and look at the AI story. Let's start with the market size.

A few few months ago, I valued SpaceX as part of the IPO and I looked at the prospect as well with the bankers had estimated a total addressable market of $22 trillion for SpaceX at lei business in SpaceX. Now total markets um clearly matter in what a business value is. But the total addressable market or TAM the acronym has become a gaming tool in the hands of founders, venture capitalists and bankers because they've realized that putting a high number out there dazzles investors into attaching a high price. At the time that I saw the SpaceX prospects, I called the 22 trillion more hallucination than estimate. And I'll explain in a few minutes why I think it's a hallucination. But I think any discussion of AI has has to start with a question. How big is the market going to be for AI? Now given that I'm a novice in the space, I might be completely missing it. So help me along. Let's see where we are right now. If you look at

the three major LLMs, OpenAI, Anthropic, and XIA, and you give them the most generous estimate of revenues you get, which is called the annualized revenue run rate, where you take the revenues in the last week or the last month and you annualize it. The reason I call it the [clears throat] most upbeat and optimistic estimate you can get is you're taking the most recent period with the revenue. So, maximize say what if the next year looks like that. Let's start with the good news. If the if you look at the annualized revenue run rates across the board at LLMs, they're all going up. In fact, just a couple of days ago, Anthropic reported an ARR of 65 billion through July of 2026. And as you look across the months, the numbers have risen. Open AI and now is second. It's rising a little slower. Grock is even slower. But basically, you can see the good news here is that the revenues are growing fast. And for many people, that's all you need. Revenues are growing 80% a year. Therefore, I will pay two trillion. I'm not ready to go

there. Even if you add up these upbeat estimates collectively, the three LLMs, you'd be lucky to get to 120, 130, 150 billion. And if you even if you add all of the other companies that provide AI products and services, remember the companies that sell into the AI architecture are not even players here. These are players who sell AI products and services. The highest estimates I have found out there 250 billion and that's stretching the limits of what these revenues are. You say 250 billion is a lot of money. It is but not if you're talking about a market pricing of 2 trillion or investing 2 trillion in capex. Right? So right now the business is growing. That's a good news but it's small. So really the question we face is what will that 250 billion will it become a trillion, 2 trillion, 5 trillion, 10 trillion or 22 trillion if the SpaceX bankers are right. Let's set some limits.

The story of AI is that companies will use AI products and services to make themselves more efficient. It'll be an operating expense that replaces something or becomes part of operating expenses. So let's start by setting a limit. If you look collectively at what every publicly traded company in the world spent on operating expenses in 2025, that number was 64.9 trillion. 64, that's a lot of money, but that is the absolute cap. Even if AI replaced every single operating expense, you replace with an AI product, that's the absolute limit. And in fact, in this table, you can already see where AI's prime targets will be in terms of sectors.

The group that has the largest operating expenses and presumably the biggest target should be industrials. Technology is only 9.2%. Industrials are 18.75%. Consumer discretionary is 16%. Now if you look across regions the part of the world which accounts for the biggest se segment of is the United States. 31% of operating expenses of all companies are in the United States followed by the EU region and but China is in there as well. So regionally that's where AI potentially could have its biggest market. So let's start with that. These are all operating expenses.

But remember operating expenses include not just employees but raw materials. And as far as I can tell, AI cannot replace the rubber you need for making tires or the chemicals you need to make fertilizers, the wheat you need to make food. So really, we should be focusing on how much of these expenses companies spend on employees, right? Cuz that's really AI's target where it offers either a tool to employees or replaces employees. Now, that statistic is tougher to get from financial statements because not all companies report employee expenses. But there's another way you can get to this number. In the US, a Federal Reserve keeps track of how much money companies spend on employee compensation. So there's a graph that shows you broken down by, you know, very broadly into into companies and goods, companies and services and government.

And the total compensation in 2025 across all of across all business in the US, public and private was 12.96 trillion. Think of that as the total compensation of employees. And if you replace every single employee in the US with an AI agent, that's going to be the absolute ceiling, right? Cuz presumably these agents are saving you money, so it should be lower. See, what about rest the rest of the world? There's no easy source for the compensation of the rest of the world. The EU, it's roughly 9 trillion in compensation. But as you get to markets like India where people where where the you know not only economy is smaller but compensation is low you're adding collectively maybe 25 to 26 trillion in total compensation. So set that out there as the absolute high value of replacing every single employee with AI. But now let's look at what portion of the 26 trillion is really going to be a market. One of the big debates about AI is it's going to be a

tool that augments productivity or is it going to replace employees? Is it going to happen quickly or is it going to happen slowly? Now earlier this year [clears throat] I wrote I did a session around the Citrini report that looked at what would happen if AI's disruption was decisive happen quickly and replace people and the bottom line of that report was while this is good news for AI it's bad news for the economy because you'll have so much unemployment and so much income leaving the economy that you're going to go into a steep recession or maybe even to depression. Now, I'm going to take the speed of disruption on one axis, the magnitude of disruption the other, and break out how big the AI market is going to be based on what it does. The biggest AI market is going to be if it replaces employees. Here's why. If AI is a tool, it's an added expense. So, if your expenses for employees are 26 trillion, you can't spend another 20 trillion in AI because that'll make you that'll

render you bankrupt. So if you use it as a tool, it's not that you're not replacing employees. You're not replacing them now, but you will hire fewer employees in the future. It's a smaller market if it's a tool rather than a replacement. As for the speed of disruption, there's an incentive for existing AI companies to try to speed this process up because if you can speed up disruption, it'll allow the current players to have an advantage, the entropics and open AI because it's happening too soon for a new entrant to come in, it'll be terrible for society because all those displaced laborers will have no place to go in the near term. So what I've listed out here the consequences in the matrix of what the products will be and then for to replace employees the products have to be highowered highriced AI whereas for tools they can be lower price there you know they can be ma mass market and I've looked at the potential market much bigger for replacement than for a tool and the winners in each scenario and the

side cost for society you get a chance take a look because what's good news for AI I can be bad news for the rest of us. Second, we're still wrestling with how much these AI agents will be priced at. I mean, right now there are some, you know, Claude Fable who estimate cost thousands of dollars per hour to run. And if you think about the premium AI products as [clears throat] being highriced, they can replace only highriced employees. You're not replacing a $30,000 employee with an AI agent that cost you $6,000 a day to run. See, here's what I did. I took the income that US workers earn. Look at what percentage is earned by quintile. And if you look at the highest quintile of workers, they account for about 52% of wages. So the 26 trillion, this quintile is the one that's most exposed to being displaced by AI. So already you

can see the market size shrinking to not from 26 trillion to a far smaller number. But there's more shrinking to come. As you look across professions, you can already start to see the professions that are going to be most susceptible to AI disruption. So you know these are professions where you do a lot of rulebased work. It's easy to check your output. Software engineers for instance are being displaced. You can see why. But jobs that require that you interact with people where their output is fuzzier, more difficult will be less vulnerable.

So in fact, this is I'm not claiming this graph is right, but it came out of one attempt to look at what kinds of businesses are most likely to or works is most likely to be replaced by AI and you can see it varies across businesses. I mean it seems I mean this might be a generalization but white collar is more likely to be replaced by than blue collar and among white collar white collar workers who do kind of routine things over and over even though it might have required specialization and expertise at one point in time are more likely to replace bankans and consultants watch out. So that shrinks the market further. And finally, if you remember the operating expense by geography, there's another component to geography, which is you look at high income earners as your potential targets for AI disruption. It varies widely by geography. North America, USA in particular, 10% of all workers make more than $100,000. But that statistic drops off as you look at the rest of the world. It's lower in Europe. It's lowest

in Africa and Asia. Australia jumps again but only 4%. Already you can look at AI disruption. If the AI products are premium products are replacing workers more likely to happen in the US, Canada and parts of Europe than in much of Asia or Latin America. So here's what's going to determine what you end up with. And I'm going to emphasize you because each of us will have to take our own shots at what the AI market is going to look like for the total addressable market for AI products and services.

If you believe it's going to be you know the current market is only but as I said about 250 billion with optimistic estimates. If you believe it's going to become mostly tools for the mass market to use in a small subset of businesses, a banking tool, a consulting tool, market's going to be tiny, less than a trillion. If you think it's an AI tool that will be used in many businesses and it'll replace highly paid workers in in a small subset in some geographies, the market gets bigger, 2 to 5 trillion depending on how if you think it's going to replace skilled workers across many industries and many geographies, not just a subset, the market gets bigger, 5 to 10 trillion. And if your total addressable market is more than 10 trillion, you see AI whether you tell me or not replacing most workers across many industries and across lots of geographies played out in terms of what that means in terms of unemployment the economy. But already you can see why three people looking at this market can

come up with three estimates. And right now I don't have a problem with people making different estimates. But people make estimates but they don't give you their reasoning. I think when somebody says the market for X AI products is 22 trillion as the bankers did that prospect is there needs to be push back. Tell me the story for AI tool replacement highest paid workers or all workers just in a in some sectors or all sectors mostly US or global that you're telling that allows you to come up with the 22 trillion.

The second stop I'm going to make in building AI as a business is industry economics. No, after all, you can have a big market, but big markets don't always translate into big revenues. Big revenues don't translate into big profits. Big profits don't translate into big value. Lots of bridges to cross. So, I'm going to start by looking at business models. And you can see we're in the phase of AI. That's why I'm calling it the barit moment where different AI companies are trying different models. and as well as the unit economics and economies of scale that at least the moment we see in this market let's start with business models when AI started there were a lot of subscription models chat JPT subscriptions anthropic offered subscriptions one trend we're noticing across the board is we're starting to see a shift away from subscription models to usage models and there's a good reason for that unlike much of technology where the marginal cost of that additional unit is close to zero. So you can just scale up.

AI products and services cost money. So giving people subscription model and not limiting the use of the model will make them cost centers rather than profit centers. And every single AI company has discovered it. So if subscription models persist, it's with caps on usage. But increasingly you see a shift away from Anthropic is furthest along because it's the most enterprise focused of the companies and you can see that in 2026 85% of entropic revenues came from usagebased models rather than subscription models. Open AI has always been tilted more towards subscription models because of chat pt's popularity but there too see you're seeing a drop off. Grock has you know is still I'm not sure where it tries to fit in here. It's still trying to find its way has also seen. So all of the models are shifting away from subscription models towards usage models. But here's what I see as the end. It's not that subscription models will go away but in the mass

market with certain types of users maybe individual subscription models will persist with caps on usage but increasingly businesses I see shifting away to usage based models. Second stop, open versus closed models. Again, as a novice, here's what I see as a distinction. When you create an open AI model, your clients can mess with the model. They can add data. They can modify the model to reflect the users. Great, right? Adaptable models. Closed model, you build the model and your clients really have to take the model as is. Now, this debate about open versus closed models has multiple factors going in. So it's kind of a messy debate because it's about how much power is concentrated in this business. Closed models give the AI companies more power.

Open models presumably spread it out. How much privacy here? Again, open models require less sharing of private data. And how safe where turns out open models might be easier to there's no slam dunk answer here, but from a business perspective, it matters. Closed models give companies more pricing power. They control the model. less stickier in the in the sense your clients can't live without you. From a business perspective, especially in the premium pricing segment, I can see why companies want closed mod the companies that make these models want closed models now and but I think open models will again it'll depend on what segment of the market in the mass market you're more likely to see open models and you're more likely to see different ways of presenting models. But I think this debate again is going to be a business debate. Let's talk about unit economics.

What do unit economics measure? What it cost you to make that extra unit? I'll start with some good news and then give you the bad news. The way to measure what it costs to provide AI is the the currency of AI usage is an AI token. The cost of an AI token has dropped pretty dramatically over the last three or four years by 80 90%. You're saying good news, right? But here's the bad news. The bad news is the AI products you're getting that you're accessing to are not getting cheaper. You're saying how come the tokens are getting cheaper. How come? Because the models we're using are getting more powerful, more output driven. And the reason output driven matters is input tokens cost a little a lot less than output tokens. So the more you're asking your model to write reports and produce output, the more tokens it's using. The net effect is AI the cost of providing AI products and services is not getting lower even though AI tokens are getting cheaper but there's a divergence coming because you

don't have to build more powerful models. If you become a mass market AI company you might say look I'm going to keep my models basic and the cost of AI products will go down in that market. In premium markets, it's going to be a race because AI tokens can get cheaper, but you're building more and more powerful models. The net effect might be more expensive models. And I think if I were cautioning AI companies, it's if you build models that are overpowerful, power that you don't need, you're costing your clients too much. And you're going to get bypassed by somebody who goes to those same clients, offers something they need. So as you look at the banking market, the consulting market, the software market, if I were advising entropic and open AAI, my advice is don't build models that are any more powerful than they need to be.

There's no point powering up a model five times more than you need to because it's just going to make it more expensive. The third part of building a business is not only need do you need to make money, you need to be able to continue to make money. In other words, you got to tell me what your modes and competitive advantages are. And this is a generic table of what kind of competitive advantage you can find. And this stre stretches across. In consumer product companies, it can be brand name. In a lot of technology companies, it's switching cost. The cost of switching into your product is low. The cost of switching out is high. It could be networking benefits, which is as you get bigger, it gets easier to get bigger in some businesses, right? Sharing is a classic example. You know, cost advantages. You might have an advantage on cost in which case allows you to charge a lower price and sell more of the same price and make more or you can get legal protection against competition. As you look at the AI business unfold and you say which of these different competitive advantage is going to kick in, I think it's going to depend in mass market AI where the

products and services can be pretty standardized and basic. It's going to be a cost advantage. perhaps the largest companies scale or because you have proprietary access to data that allows you to supply the products, you have an advantage in premium AI where products are customized and powerful. I think the winners are going to be the companies that can not only handcraft these products, they have the technology, know how to do it, but are able to bring the cost down as quickly, much more quickly than the rest of their competition. And since it's built around client data, the more you can entwine your model with client data, the stickier you become.

It's tough for your clients to let you go because you are so much you you have so much of your data built into the models. No. So mass market can legal protection help you? There are some parts of the world where I think AI companies going to prosper because the country they're in wants to protect the AI product and service in that country and it's going to keep the competition out. It's going to be interesting to see the modes play out. Can people be an advantage? You know, some of you might have read the story of Jeff Dean, Google's chief scientist. He was actually chief scientist at DeepMind, which is Google's AI entity. And when he left the company a couple of weeks ago, um, Alphabet, Google's parent company, lost almost 5.4%. I mean, that's almost hundred billion dollars because one person left. Every day you wake up to another news story about AI firms raiding a university, hiring computer science and technology talent, AI econ focused economists, even a few philosophers and we'll talk about why that might be. Right now people matter

and I'm not being cynical when I say as the and that's an indic indication of how young this industry is. But as the industry matures, it's almost a given that it's not that people will not matter, but they will matter less than they do right now. You say, "What about brand name? What does brand name mean? Would you be will willing to pay a higher price for a clawed agent or an AI agent from a different company?" Now, I think that the answer is still being worked out because in addition to all of the standard business reasons, more product power, pricing, reliability, trust enters the equation. You're saying, "What's trust got to do with it?" Cuz when you let an AI company enter the game, you're trusting them with your data, trusting them with the inner secrets of your enterprise. And no, so if you don't trust them, you're not going to use them. So it shouldn't come as a surprise that AI companies are competing in the virtue space. That said though, you can call me cynical for saying so, but when I hear Dario or Sam Alman wax eloquent about how they're

going to make the world a better place and make their products safe, I mean, I'm reminded of Shakespeare. So, thou protests too much. I mean, you know, ultimately, no matter what these companies say or how many philosophers they hire, their actions speak louder than words. So, they will be judged on how they behave, not on what they say. So with that said let's talk about the constraints and limits that are going to come on AI and you can see the roots of the push back you know starting already why is there already a push back against I can think of four reasons why a AI is evoking the kind of negative reaction it is the first is unlike previous iterations of technology AI has a footprint a footprint that affects people in the form of huge capital investments are re in real estate. A typical data center can be the as large as 100 150 football fields. And while you can sell the commu community

on the economic benefits of data centers, it is also true that it alters the lives and the lifestyles of the people who live there, not to mention the power and the water usage, which also puts a drain on natural resources. All of which I think create a push back. Now uh there's a data privacy issue after 20 years of social media companies invading our privacy by using the data that we have contributed. So we have a we have a role in this process where we share data. We worry about AI companies getting data on us and using it against us. As I said the best case scenarios for AI will also be scenarios where people lose jobs. Hundreds of thousands of people many of them well paid. those are not good for I mean those job losses will be not good for the economy and not good for society and that worries people as well and finally we live in an age where income inequality is one of the key topics in politics and economics and

there's a sense that AI if it rises will increase the inequality that's fed into it SpaceX went public it created dozens of centillionaires people with more than hund00 million and anthropic and open AI do it I'm sure those new stories will come out as well. And all of that is going to feed into the fairness question. Why does this matter if you're investing in AI as a business? Because it's going to manifest in push back. Data centers are going to get more difficult, more expensive to build. As concerns about data mount and there are scandals that pop up here and there, you are going to get restrictions in the use of data which is going to increase the cost of acquiring and protecting data.

The worries about AI agents replacing people will show up in systems stepping in and protecting people's jobs. They're requiring these AI companies to pay for the employees that get displaced. And finally, I you know, you might you and I might think billionaire, you know, taxing people on wealth is neither fair nor efficient, but that movement is just going to get added ammunition if AI is as successful as people claim claim it to be. So we have to build these restrictions in. So all of that is in the macro. You're saying why should I care if I'm interested in investing in anthropic or I'm investing in meta and want to know whether their capex pays off. No [snorts] I think to get to the micro there are two ways you can approach this. First is you can look at the pricing of these companies or the capex are making and ask what will need to happen in terms of revenues and income to justify that upfront investment. The other is to build a full-fledged valuation of these companies where you look at their business models and their unity

economics. It's it's it it sounds like hard work and it will be but to build up to a valuation. So let's take the first reverse engineering. We set up the process. I spent a lot of my time in intrinsic valuation. Intrinsic valuation if you cut it down to basics. Here's what we do. We take a company. We estimate what it will generate as cash flows in the future. Cash flows come from first projecting what you would get as revenues, subtracting out your expenses to get to earnings, and then netting out the reinvestment you make to get future growth to get to cash flows.

You project those cash flows for as long as you can. Then you stop. And because you can't ignore cash flows beyond that, you put closure by assuming your company becomes mature and use the cash flows in that phase to get a terminal value, the value of your company at the end of your closure period. You discount them all back at a risk adjusted discount rate. you get the value today. Now, in reverse engineering, here's what you do. You take the current market price or the capex you're making as a given. And then you reverse the process. You ask yourself, how much will I need as revenues by the time I get to steady state given my margins and given my risk and reinvestment needs to justify what I pay today? It's a break even point. Now when I tried this on entropic with a $2 trillion pricing which is what's being rumored now I you know my and I give them a a healthy after tax operating margin 30% and treat them as above average risk cost of capital of 10%.

They would need to generate about 1.2 trillion in revenues if you have to wait 10 years for this business to mature or two trillion if you have to wait 15 years. That weight can come not just because the technology is not there but because of regulatory push back. two trillion. Now think of that because when you see people getting excited about you know annual revenue run rates being 65 billion 65 billion to 1.2 trillion or 2 trillion is a big gap to make up. If you look at AI companies in the aggregate add up all of the market pricing and I'm being conservative when I give that total number 5 trillion probably higher than that. It turns out that if you have to wait 10 years and you give them lower margins collectively because some of these companies are lower margin companies, you'll need about five trillion in revenues in year 10 to break even if you have to wait 10 years or 8 trillion if you wait 15 years. I mean I'm not suggesting you know again bubbles high price basically I'm saying AI companies right now are priced for

immense success in the business and I put a number on what that success will have to look like. Of course, the other approach is to value a company from the bottom up. What will that require? Making choices, making assumptions about what kind whether the company is going to focus on the mass market or the premium market, what the unit economics will be in that market, what kind of competitive advantages do they bring to the game, what investments they will need to deliver that growth. And some of that you're going to look at what they've already invested. Companies that already invested might not have to invest much more or companies that invest more efficiently. Remember the deepseek disruption that happened a couple of years ago where Chinese companies showed that they could deliver many core AI products and services without investing in expensive data centers or Nvidia chips. And finally, the regulatory constraints. So as entropic and open AI file their prospect is I I look forward to taking those companies through this process and valuing them but we're not there yet. So what's the bottom line? Now, I told you

at the start of the session that I was writing this for myself. I was writing for this for myself because I find myself put in a dozen different directions looking at all of these different AI debates on different dimensions. I desperately needed a framework to bring all these discussions together to understand why they matter or whether they matter in the first place. Now, it's a it's my it's this is a first try and I'm sure I haven't got the framework right. There are holes in it and I hope you point them out if I miss something but I'm cleareyed about what the framework cannot do. It cannot tell me whe the total addressable market for AI is 10 trillion or 8 trillion. It cannot tell me whether anthropic is priced right or priced wrong at 2 trillion. But here's what it can tell me. It can give me tell me how I should think about coming up with the total addressable market for anthropic. It can tell me that a $22 trillion market put into a prospector is is basically made

up because it doesn't fit the facts. And it also tells me that justifying paying 2 trillion for anthropic because it's a it's a annualized revenue run rate rose 80% in the last 3 months or 6 months is neither here nor there. As you work through this framework, I'm sure you will find disagreements with me and I hope you do because that's the point of a framework is not that we all arrive at con conviction. But here's the one thing that I hope you take away from this framework. There's lots we don't know.

Anybody who claims has conviction that they know what's coming in the air is either ignorant or arrogant. I mean, if you remember a session I did a couple of weeks ago on Leo Ashen Brener and his situational awareness hedge fund. I take issue with them on his view that AI would be disruptive, massively disruptive, happen quickly. I took issue with him on the conviction he felt in that view that he was able to go out and borrow immense amounts of money. This is not a space where you can have conviction yet. You can make your best judgment. You better but you need to hedge your bets. You need to spread your bets. I hope you found the session useful in creating that framework for you to think through your own answers to AI. And thank you very much for listening.