Ex-buyside/sellside. Spent years writing research nobody read on stocks everybody owned. Now writing research people actually want on stocks nobody covers.
Very cool insights: Agree 100%. One thing I would add is this: many/most investment professionals (perhaps as a result of learning efficient market theory or some such) may think that there is too much information - but I think this is false in a crucial sense...yes, we have never had more information available to us - and yes, the market is extremely good at processing that information...but in another sense it is false: there is one kind of information that we have very little of in the market (or very little available to most people)...and that is procedural knowledge...knowledge of 'how-to', rather than propositional or representational knowledge (graphs, pictures, data, theory etc.)...So what is procedural knowledge and why could it be a crucial missing ingredient? Procedural is the difference between knowing how to hit a forehand and being able to describe it or draw a picture of it...it is the difference between running a company and learning the theory of how to run a company...it is tacit...
A picture may be worth a thousand words - and there is a good reason why there is this saying (related to information theory for those who are interested) and its implication - that a thousand and one words many be worth more than a picture....the challenge and opportunity of tacit/how-to/procedural knowledge is that it is difficult to - or seems incredibly hard to - turn into a recipe or a picture...not impossible (as humans do it, and we are "computers"), but it seems that whereas a picture may require a 1000 words, know-HOW - would require a thousand pictures or a million words....
In short (lol, I see the reply turned out much longer than I initially planned!): (So, in long): I think we have way too little of this procedural knowledge available to AI's at the moment (they are trained on propositional and representational knowledge - and tangentially by procedural knowledge as fragments of our know-how seep through into our language and representations)...To gain more procedural knowledge, and hence the possibility to map human nature and behaviour much better - would require AI's to gain access to procedural knowledge in a far more effective way (how? AI's need more - and different kinds of sensors: just like a blind person won't know what red is because they sadly lack the right sensors for that kind of knowledge, AI's largely lack the kind of sensors that can "see" procedures to the extent required for their analytical and synthetic powers to create mastery...somehow, they need to see/feel/experience us acting, this final step would be a fantastic learning experience for them...one that they currently only "see" through the metaphorical and pictorial traces we make available to them through the Internet/etc.
I agree with the argument that investment judgement is largely learned.
What we call experience is an accumulation of mental models, pattern recognition, causal understanding and feedback from previous decisions.
An experienced investor has seen businesses succeed and fail, navigated different economic environments and developed an intuition for which pieces of information matter.
There is no obvious reason why machines couldn’t learn many of these things. In fact, they have some considerable advantages.
An investor might experience three recessions in their career. An AI system could study hundreds of historical episodes across thousands of companies, identify recurring mechanisms and test whether those relationships hold elsewhere.
But I think there’s an important distinction we can add to this debate.
Learning patterns is not necessarily the same as understanding the underlying mechanisms.
Financial markets are particularly difficult environments for machine learning.
Relationships are constantly changing. Feedback is noisy and delayed. Historical correlations break down. And the behaviour of market participants changes in response to the very signals they are trying to exploit.
A machine might learn that declining restaurant traffic tends to precede earnings downgrades.
But understanding whether that decline reflects temporary weather disruption, structural brand deterioration, competitive displacement or a deliberate pricing strategy requires something more than pattern recognition.
The initial evidence might look remarkably similar. The eventual investment outcomes could be entirely different.
This is where causal reasoning, competing hypotheses and an understanding of how businesses actually operate become so important.
The real opportunity is therefore not just to build AI that learns from historical investment outcomes, but AI that develops and continually challenges its own understanding of the world.
Imagine a system that connects company filings, alternative data, industry dynamics, management commentary, historical precedents and market expectations into an interconnected body of knowledge.
It develops hypotheses, identifies relationships, investigates anomalies and builds competing explanations for what is happening.
It makes forecasts, tracks the outcomes and examines where its reasoning was wrong.
Importantly, it doesn’t simply adjust a forecast when it misses. It investigates whether the underlying mental model was flawed, whether the evidence was misleading or whether the world itself has changed.
Over time, it develops a more sophisticated understanding of businesses, industries and the relationships between them.
This is fundamentally different from automating the existing research process.
There is another challenge, however: investment outcomes are not clean feedback signals.
A good decision can lose money. A poor decision can make money. A forecast can be wrong for entirely legitimate reasons in a probabilistic world.
The system must therefore learn to distinguish between errors in reasoning, errors in forecasting and the randomness inherent in the outcome.
Otherwise, it risks optimising for historical patterns that have little predictive value.
And this leads to what I think will ultimately define investment edge.
Human investment methodologies evolved around human limitations.
We have limited attention, memory and processing capacity. We cannot investigate every company, connect every dataset or test every possible explanation.
Our research processes, organisational structures and investment frameworks have all evolved around those constraints.
AI changes that equation.
The question is no longer just whether AI can learn to invest like us. It’s whether we can build systems that learn how to invest in ways we haven’t yet imagined.
“AI won't replace human judgement in investing”
I think a lot of investors are betting their careers on that holding up.
Here’s why I think it breaks down, and where investment edge moves as a result.
“AI won't replace human judgement in investing”
I think a lot of investors are betting their careers on that holding up.
Here’s why I think it breaks down, and where investment edge moves as a result.
Underrated observation. Anyone who has spent time in the industry knows there’s often a meaningful difference between an analyst’s published view and their actual assessment of a business.
Grab a beer with a sell-side analyst and you’ll often get a much more candid version of their thinking.
The published note is one output of the research process, filtered through commercial, reputational and regulatory considerations.
The real value is often in the conversation: what they privately question, which management responses they don’t buy, what they’re hearing across their industry network and where they think consensus is getting it wrong.
A price target is easy to read. Understanding the analyst behind it is considerably more valuable.
If you’re only reading published sell-side research and not getting the analyst on the phone, you’re getting half the picture at best. Sell-side analysts have every incentive to keep bearish views off paper. You risk losing management access and jeopardizing your firm’s capital markets business. That’s the reality of the business. The published research, including the estimates, are the sanitized version. Few seem to understand this.
If you’re only reading published sell-side research and not getting the analyst on the phone, you’re getting half the picture at best. Sell-side analysts have every incentive to keep bearish views off paper. You risk losing management access and jeopardizing your firm’s capital markets business. That’s the reality of the business. The published research, including the estimates, are the sanitized version. Few seem to understand this.
This is the best thing I've read on process in a while. On the "ascendant and dying" point, I think the answer is fairly simple: the style is ascendant in AUM and dying in edge.
The pod platforms are still raising money and their risk setup more or less forces this way of working. If you get cut at a 5% drawdown, you don't get to call the next two quarters noise. The problem is that everyone now has the same card data, the same expert calls and the same AI tools summarising the same transcripts. Getting gross margin within 100bps has become the minimum, and nobody gets paid for the minimum.
When I was on the buy side, the work that actually mattered was understanding the product. What does the business actually sell, who buys it, and why do they come back? If it's Lululemon, go to the stores, try the gear and see whether the new ranges are landing or whether people are walking into Alo or Vuori instead. With cruise lines it helps to map the whole chain, from the handful of yards that can build the ships to the private islands and onboard spend that drive the economics. With Boeing you learn quickly that the production rate is set by the weakest supplier, not the order book. Software is harder because there's nothing to touch, but you can still use it, talk to the people who live in it every day, and work out whether it's the system a customer runs the business on or a feature someone bigger will bundle.
Once you've mapped the product against the market and the other players in the value chain, you usually have a much better sense of where the industry is going in three years than any gross margin forecast will give you.
None of that means throwing the model away. "Qualitatively precise" is easy to claim and very hard to check. In 2021 lots of people had beautifully argued three-year views, and plenty were wrong in ways the numbers would have flagged if anyone had been looking. The way I'd use the model is to keep the long-term view honest. If I'm right about where value accrues, what has to show up in the numbers, and roughly when?
Write that down. Then a 100bps miss on gross margin is either evidence against what you believe or it's noise, and you actually know which.
Now I'm on the sell side I see the other end of this daily. The flow rewards the post-conference note with the estimate changes. But the analysts people still talk about years later were rarely the ones 50bps closer on a quarter.
I'm also not sure it's really a TMT thing. The product work matters just as much for a retailer or a cruise line. Software just makes it easier to skip, because there's nothing to try on.
Genuine question: there’s a (big) group of Gen X PMs who got very rich operating this model. They live in the spreadsheet, keep pressing the same checks for trend lines, talk to management twice a quarter to detect some slight change in tone or explanation of a driver, then run back to the model and tweak incremental margins or customer adds or whatever for the next two quarters.
I feel like this form of investing is both ascendant and dying at the same time.
I get that “it comes down to the numbers,” but increasingly I think getting the big annual P&L wins requires being “qualitatively precise”. By that I mean understanding the rate of change within an industry, anticipating inflections based on first principles (not today’s credit card data), and having a deeply informed view of how a business evolves through technological shifts, competitive shocks, and cycles.
These approaches obviously aren’t mutually exclusive. But I rarely find the Gen X model-obsessed set equally obsessed with understanding how the product actually works, which architecture survives change, where value accrues, how a market absorbs incremental competition, etc. Maybe this matters disproportionately in TMT.
But the guy who runs back to his model to update next quarter’s GM% based on an offhand management comment at a conference rarely seems to be the same guy who can tell you, with any real precision, what the business is going to look like in three years. And I wonder whether the second skill is actually the better way to get the next two quarters right too (or at least treat them as vol or margins of errors as noise toward a more inevitable outcome). Checks still matter. Listening for change matters. But knowing what the heck you own really matters and the credit card data or next quarters RPO within 100bps doesn’t tell you that.
Genuine question: there’s a (big) group of Gen X PMs who got very rich operating this model. They live in the spreadsheet, keep pressing the same checks for trend lines, talk to management twice a quarter to detect some slight change in tone or explanation of a driver, then run back to the model and tweak incremental margins or customer adds or whatever for the next two quarters.
I feel like this form of investing is both ascendant and dying at the same time.
I get that “it comes down to the numbers,” but increasingly I think getting the big annual P&L wins requires being “qualitatively precise”. By that I mean understanding the rate of change within an industry, anticipating inflections based on first principles (not today’s credit card data), and having a deeply informed view of how a business evolves through technological shifts, competitive shocks, and cycles.
These approaches obviously aren’t mutually exclusive. But I rarely find the Gen X model-obsessed set equally obsessed with understanding how the product actually works, which architecture survives change, where value accrues, how a market absorbs incremental competition, etc. Maybe this matters disproportionately in TMT.
But the guy who runs back to his model to update next quarter’s GM% based on an offhand management comment at a conference rarely seems to be the same guy who can tell you, with any real precision, what the business is going to look like in three years. And I wonder whether the second skill is actually the better way to get the next two quarters right too (or at least treat them as vol or margins of errors as noise toward a more inevitable outcome). Checks still matter. Listening for change matters. But knowing what the heck you own really matters and the credit card data or next quarters RPO within 100bps doesn’t tell you that.
I spent a five years on the buy side before moving across, and I can't actually remember marking anyone down for a call that didn't work. What got on my nerves was the analyst who went quiet when it was going wrong, or who'd written something so hedged you couldn't work out what they really thought. The ones I rated would ring up and say, here's where I got it wrong, here's what I'm watching now, and here's whether I still think the case works.
Everyone gets calls wrong. People remember how you handled it far more than the call itself.
The bit about management respecting the analyst who disagreed with them rings true as well. You can usually tell on a call who has done the work and who is just asking for the guidance again.
Where I'd see it slightly differently is the idea generation point. I don't think corporate access killed it so much as the money did. Over here in the UK, MiFID II took a big chunk out of research budgets, so you've got analysts covering more names with less time and fewer people around them. Access was the thing clients would still pay for, so that's where the effort went. Feels more like a symptom than the cause.
The slightly ironic part is that it makes a properly researched contrarian call worth more now, not less, because so few people have the time to do one. In smaller caps you have to be careful, as a management team can make life awkward if they don't like what you write. But it's also where a different view gets noticed quickly.
Yes. I agree 100% with each and every point in that thread. Most people think the job of a sell-sider is to get your stock calls right. Its not (although it does suck when you are wrong). Its mainly for idea generation, corporate access, and entertainment.
Increasingly though the sell-side isn't doing much idea generation, other than the telephone game. And its entirely because of corporate access.
The funny thing is, my best corporate relationships often began when I took the opposite view of management. Required real work. Had to be balanced (not just being negative for negative sake). But over time those companies came to respect me a lot more than the infinite kiss asses.
But current sellsiders, at least the younger ones, don't have the luxury of time to use that playbook. They need management events now or they wont have a job much longer.
Many people on the buyside LOVE to talk about sellsiders. Its like a neverending reality TV show (maybe even more like WWE wrestling). Who is going to get the Citigroup Softlines job now that Paul Lejuez is going to Cowen? What happened to Oliver Chen at Cowen and does that mean he won't hand out fake Chanel handbags at the ICR conference anymore? So and so's downgrade on Nike is so "stupid". So and so's upgrade of On is so "smart". You have your "heros" and your "heels" and they often flip just like Hulk Hogan did in the mind of clients. And all of this plays out in real time, all day long, in chat rooms, text chains, phone calls, meetings, etc.
Its not that different from X actually. Its just 100% public here.
Sometimes, the job of a sellsider is to be a target of ridicule so that buysiders forget their own mistakes. Call it like a weird form of therapy. Some of the highest ranked sellsiders have the worst stock calls.
One time made a bad call on a stock and was young and letting it get to me. Remember being at the Fossil investor day like 15 years ago, where a consensus top five retail/consumer buyside PM at that time pulled me aside. He said something that really lasted with me for the rest of my career: "The only difference between you and me is that everybody always knows when you get a call wrong".
Great points, but I'd go a step further.
If I had the choice, the entirety of my modelling and forecasting would live in Python. Agents are native in code, git gives you full version history for free, and every driver can be a distribution rather than a point estimate.
I'd also question how long humans keep the edge on spotting patterns in the history. Agents can iterate through filings far faster than we can, and I suspect they'll get better than us at catching margin inflections or shifts in seasonality. The learning moves from keying in numbers to interrogating the agent's reasoning.
Which is where the front end matters. I don't think the answer is Excel with extra UI bolted on. It's a clean HTML output sitting on a Python back end, where you can click into any driver, forecast or assumption.
Click on FY27 revenue growth and out pops a driver card:
1. The forecast as a distribution, not a single number
2. The path: how the range has moved over time and what moved it
3. The evidence: management guidance, industry data, alt data, all cited
4. What agrees and what disagrees
5. A probability grounded in history, e.g. how often this company has actually delivered on guidance like this
Compare that to an Excel cell that says 8% with no reference to why it was modelled that way.
Excel probably survives as the export for colleagues who want a spreadsheet. It just stops being the engine and the interface.
imo it DOESN'T die, but bear with me on this
whether excel dies or doesn't die has been topic du jour in investing for at least the last 2 years and I wanted to weigh in
for me there are 2 sides to this:
1. Excel isn't the most natural tool for agents to model in. Making them operate cell by cell feels like square peg round hole (and we've tested this too). It can add errors, there's no version history, latency and cost are also an issue. This is SO important when you're doing it across hundreds of companies.
2. Excel is still the most natural way for humans to inspect this type of work. It's familiar, you can trace assumptions and you can send it to a collague who will very quickly understand it.
it might help to start with why we model companies, in my view it's for 2 reasons:
1. Understanding the history - building the model is part of learning the business. As you key in the numbers, you spot patterns, understand seasonality, notice margins falling or cash conversion improving. Every time you spot one of these patterns you go back to primary filings or other sources to understand why which drives the learning process.
2. Marking all of you implicit thoughts explicit - once you've learned about the business through historics, you set out on a question to learn everything you can about the company and take a view. All of these implicit thoughts need to be translated into explicit forecasts so 1) you can quantify your thoughts and 2) so it can be easily communicated to colleagues or clients.
I expect both of of these jobs to survive somewhat, but how they get done likely changes:
1. Agents do the construction and maintenance - modelling absolutely still exists, but I expect us to move from point in time updates to continuous updating done by agents when relevant information arrives which changes surfaced to the human where needed.
2. More of our thinking gets represented in the model and they become far richer - that 8% revenue growth forecast is an unnecessarily precise forecast for something we know in our head is a probability weighted range of outcomes. We already use scenarios and sensitivities, but maintaining that a written down version of your expected range of outcomes for every driver is impossible/exhausting and would look very messy. If this could be maintained by agents, I'd expect forecast quality to improve significantly.
FLY IN THE OINTMENT
In this new paradigm of modelling I'm going to want to:
1. Easily see the range of outcomes for each driver
2. See the full history of how that range has changed, when it change and what changed it
3. Do all of that without having to prompt my coding agent to tell me
So there really is a UI challenge here, excel is without doubt going to be the familiar form factor but I think the best in class tools for this will add additional UI to a UI which looks like excel at it's core but it's purpose built for agents and probabilistic, constantly updating models
(yes, I realised this is a meme account)
imo it DOESN'T die, but bear with me on this
whether excel dies or doesn't die has been topic du jour in investing for at least the last 2 years and I wanted to weigh in
for me there are 2 sides to this:
1. Excel isn't the most natural tool for agents to model in. Making them operate cell by cell feels like square peg round hole (and we've tested this too). It can add errors, there's no version history, latency and cost are also an issue. This is SO important when you're doing it across hundreds of companies.
2. Excel is still the most natural way for humans to inspect this type of work. It's familiar, you can trace assumptions and you can send it to a collague who will very quickly understand it.
it might help to start with why we model companies, in my view it's for 2 reasons:
1. Understanding the history - building the model is part of learning the business. As you key in the numbers, you spot patterns, understand seasonality, notice margins falling or cash conversion improving. Every time you spot one of these patterns you go back to primary filings or other sources to understand why which drives the learning process.
2. Marking all of you implicit thoughts explicit - once you've learned about the business through historics, you set out on a question to learn everything you can about the company and take a view. All of these implicit thoughts need to be translated into explicit forecasts so 1) you can quantify your thoughts and 2) so it can be easily communicated to colleagues or clients.
I expect both of of these jobs to survive somewhat, but how they get done likely changes:
1. Agents do the construction and maintenance - modelling absolutely still exists, but I expect us to move from point in time updates to continuous updating done by agents when relevant information arrives which changes surfaced to the human where needed.
2. More of our thinking gets represented in the model and they become far richer - that 8% revenue growth forecast is an unnecessarily precise forecast for something we know in our head is a probability weighted range of outcomes. We already use scenarios and sensitivities, but maintaining that a written down version of your expected range of outcomes for every driver is impossible/exhausting and would look very messy. If this could be maintained by agents, I'd expect forecast quality to improve significantly.
FLY IN THE OINTMENT
In this new paradigm of modelling I'm going to want to:
1. Easily see the range of outcomes for each driver
2. See the full history of how that range has changed, when it change and what changed it
3. Do all of that without having to prompt my coding agent to tell me
So there really is a UI challenge here, excel is without doubt going to be the familiar form factor but I think the best in class tools for this will add additional UI to a UI which looks like excel at it's core but it's purpose built for agents and probabilistic, constantly updating models
(yes, I realised this is a meme account)
Case study #2: Dempster Mill (Buffett, 1956–63)
Second in my series rebuilding special situations from famous investor letters. Last week was Sanborn Map. This one is messier and more instructive: the first company Buffett actually had to run.
The business
Dempster made windmills, water systems and farm equipment in Beatrice, Nebraska. Static sales, slow-moving inventory, almost no profit on the capital tied up.
The numbers
In 1956 the stock was $18. Book value was $72 a share, and net current assets (cash, receivables and inventory less every liability) were $50.
You were paying 36 cents per dollar of net current assets and getting the factory for free.
Five years of doing nothing
The stock went nowhere, so he kept buying. He joined the board. By mid-1961 he had about 30%.
In August and September 1961 he bought big blocks at $30.25 and tendered at the same price, taking the partnership above 70%. Average cost was around $28.
A stock that does nothing for years is a problem for most investors. For Buffett it was the opportunity, because it let him keep buying until he controlled the outcome.
What it was actually worth
Once you control a company, the share price stops mattering. What matters is what the assets fetch.
Buffett marked every line to a quick-sale value: receivables at 85%, inventory at 60%, plant at auction value. That gave $35 a share against his ~$28 cost.
Inventory was over 60% of book assets. I ran the numbers: it only needed to recover about 50 cents on the dollar for him to break even.
The management problem
He gave the existing team six months. Nothing happened. So in April 1962, on Munger's recommendation, he hired Harry Bottle, a turnaround operator who had never run an implement company.
In under two years Bottle:
1. Cut inventory from $4.2m to under $0.9m
2. Halved admin and selling costs to $75k a month
3. Cut factory overhead from $6 to $4.50 per labour hour
4. Closed five loss-making branches
5. Raised prices on repair parts. Farmers with a Dempster windmill had nowhere else to go, so it added ~$200k of profit with almost no lost volume
The admin cuts and parts repricing alone are worth about $1.1m a year pre-tax, or $17.70 a share, on a stock bought at $28.
Factory into portfolio
Every dollar Bottle freed up, Buffett put into undervalued stocks. Cash and securities went from about 4% of adjusted assets to nearly half.
Value per share using the same haircuts: Nov 1961: $35 Nov 1962: $51 Mid-1963: $65
My favourite detail: the partnership owned 71.7% of Dempster at a cost of $1.26m. By June 1963 Dempster's safe deposit box held $2.03m of securities. The partnership's share of that box alone was worth more than it paid for the whole stake.
The exit
Bottle restored earnings so fast the tax losses ran out early, so it was either de-incorporate or sell. In late 1963 the operating business was sold for roughly book value. With the portfolio gains, the partnership realised about $80 a share.
That's ~$2.3m profit on $1.26m, 2.8x the money, or roughly 54% a year from taking control (flattering, as it ignores the five years of accumulation).
What I'm taking from it
1. Buy at a big enough discount to liquid assets that heavy haircuts still leave you covered.
2. A stock doing nothing is fine if you're willing to keep buying.
3. The value was trapped in working capital. Releasing it mattered more than growth.
4. Management was the binding constraint. Same assets, different operator.
5. An installed base needing spare parts is a small monopoly.
6. Know how you get out. Tax decided the exit, not the share price.
Sanborn's value was already liquid and needed a structure to release it.
Dempster's was stuck in a warehouse and needed someone to get it out.
Next week: Northern Pipeline, Ben Graham's 1920s proxy fight.
Good post and I agree with most of it. A few things I'd add from the other side of the table.
The circular revenue point is right, but every capex cycle looks like this early on. One firm's spend is another firm's sales until end demand turns up and validates it, or doesn't. What I watch more closely is where the money is coming from. For most of this build-out the big spenders paid for it out of operating cash flow, which is a genuine difference from the late 90s telecom boom that ran on debt. That's changing. More of the marginal dollar now comes from bond issuance, private credit vehicles, leveraged neoclouds and vendor financing, where the chip supplier invests in a customer who then buys its chips. When the funding source changes, the risk changes, even if the revenue line looks identical.
The depreciation point is the one I'd underline twice. When the fibre boom blew up, the cable stayed in the ground and ended up carrying Google and Netflix traffic for a fraction of what it cost to lay. Investors lost their shirts but everyone else got cheap infrastructure. GPUs don't come with that consolation prize. They're economically obsolete within a few years, so an overbuild doesn't become cheap capacity for the next generation, it becomes a write-down. Which is why I pay a lot of attention to the useful life assumptions sitting behind reported earnings. Stretching server lives from four years to six flatters profits more than most people realise.
Where I'd push back a bit: big share price swings don't automatically mean the market is being silly. For long duration assets, a small change in the odds of a winner-takes-most outcome should move the value a lot. A 30% move with nothing new in the quarter can be perfectly rational. The question I ask is whether the move came with new evidence on what end customers will actually pay, or just a new capex number.
The last paragraph is the bit most people will skim past, and I think it matters most. The value capture question isn't confined to chips and data centres. It hits every business that sits on top of the internet today. Price comparison, reviews, travel booking, marketplaces. For them AI is either a big margin tailwind (cheaper service, cheaper content, cheaper engineering) or the thing that competes away the moat once agents start doing the searching and comparing for people. Probably both, depending on the company. A lot of money will be made and lost there over the next decade, and it gets a fraction of the attention of the next Nvidia print.
Plenty of people will be right about the technology. Far fewer will be right about who ends up keeping the money.
My thoughts on AI 👇
Frequently the story about AI seems to change completely. First the AI boom is supposedly over, then it’s just getting started. Meanwhile, stocks swing 20%, 30%, sometimes 50% in either direction, often after a single earnings report. Yet when you look at the underlying businesses, very little has actually changed in those few weeks. What changed was people’s expectations, and the market has a remarkable habit of treating changing expectations as new facts.
We’ve seen this before with railroads, electricity, the internet and nearly every revolutionary technology. People correctly understood that the world was about to change, but that didn’t mean they knew who would make the money. Being right about a technology and being right about an investment are two very different things. History is full of extraordinary innovations that produced disappointing returns for the investors who financed them, often because competition handed most of the benefit to customers.
And I don’t believe this time will be any different. Everyone has their favorite AI companies, the ones they believe are best positioned to dominate the future. And perhaps they’re right. But unfortunately, investing doesn’t work quite that simply. A company that looks absolutely extraordinary today can look like a terrible business five years from now, for reasons that even the smartest investors couldn’t possibly have anticipated. That’s the nature of capitalism. Competition, innovation and changing economics have a way of making fools out of even the most intelligent people. I think recognizing that uncertainty and occasionally eating a slice of humble pie is one of the most valuable things an investor can do.
Here is something about this moment that I think gets too little attention. In AI today, one company’s revenue is very often another company’s expense. When a large technology company spends billions on chips and data centers, that spending shows up as exciting revenue growth for the chipmaker and the cloud provider. The market celebrates the revenue. But for that dollar to be truly profitable for everyone, someone at the end of the chain, a business or a person, has to pay enough for AI to justify what everyone in the middle spent building it. Until that happens, much of what we call AI revenue is really AI investment wearing a different label.
So I think the most important questions are still largely unanswered. Who is paying for all this computing power, and is that money coming from real profits or from capital raised? How much of today’s growth comes from end customers, and how much is money circulating between companies that are simultaneously investing in, buying from and financing one another? And will the enormous investments being made today earn attractive returns before the chips inside these data centers are worn out or overtaken by new ones?
These questions will decide who wins and who loses. A stock can rise 40% while its long term economics deteriorate, just as it can fall 40% while the business becomes more valuable. Neither move tells you much about what the company will earn ten years from now. The market changes its opinion constantly, but the economics of a business change slowly.
I believe AI will transform the world, but that alone tells me very little about which stocks will be good investments. The internet changed virtually everything, yet many of the companies that built its infrastructure never earned an adequate return on the money they invested. The winners weren’t necessarily the ones that spent the most, but the ones that eventually captured the most value. And some of tomorrow’s biggest winners may be companies we’re barely paying attention to today.
So the next time AI stocks soar or collapse, it’s worth asking yourself two questions. Did the economics of these businesses actually change, or did the market simply change its mind? And how much of what I believe about the future do I actually know?
🌹
My thoughts on AI 👇
Frequently the story about AI seems to change completely. First the AI boom is supposedly over, then it’s just getting started. Meanwhile, stocks swing 20%, 30%, sometimes 50% in either direction, often after a single earnings report. Yet when you look at the underlying businesses, very little has actually changed in those few weeks. What changed was people’s expectations, and the market has a remarkable habit of treating changing expectations as new facts.
We’ve seen this before with railroads, electricity, the internet and nearly every revolutionary technology. People correctly understood that the world was about to change, but that didn’t mean they knew who would make the money. Being right about a technology and being right about an investment are two very different things. History is full of extraordinary innovations that produced disappointing returns for the investors who financed them, often because competition handed most of the benefit to customers.
And I don’t believe this time will be any different. Everyone has their favorite AI companies, the ones they believe are best positioned to dominate the future. And perhaps they’re right. But unfortunately, investing doesn’t work quite that simply. A company that looks absolutely extraordinary today can look like a terrible business five years from now, for reasons that even the smartest investors couldn’t possibly have anticipated. That’s the nature of capitalism. Competition, innovation and changing economics have a way of making fools out of even the most intelligent people. I think recognizing that uncertainty and occasionally eating a slice of humble pie is one of the most valuable things an investor can do.
Here is something about this moment that I think gets too little attention. In AI today, one company’s revenue is very often another company’s expense. When a large technology company spends billions on chips and data centers, that spending shows up as exciting revenue growth for the chipmaker and the cloud provider. The market celebrates the revenue. But for that dollar to be truly profitable for everyone, someone at the end of the chain, a business or a person, has to pay enough for AI to justify what everyone in the middle spent building it. Until that happens, much of what we call AI revenue is really AI investment wearing a different label.
So I think the most important questions are still largely unanswered. Who is paying for all this computing power, and is that money coming from real profits or from capital raised? How much of today’s growth comes from end customers, and how much is money circulating between companies that are simultaneously investing in, buying from and financing one another? And will the enormous investments being made today earn attractive returns before the chips inside these data centers are worn out or overtaken by new ones?
These questions will decide who wins and who loses. A stock can rise 40% while its long term economics deteriorate, just as it can fall 40% while the business becomes more valuable. Neither move tells you much about what the company will earn ten years from now. The market changes its opinion constantly, but the economics of a business change slowly.
I believe AI will transform the world, but that alone tells me very little about which stocks will be good investments. The internet changed virtually everything, yet many of the companies that built its infrastructure never earned an adequate return on the money they invested. The winners weren’t necessarily the ones that spent the most, but the ones that eventually captured the most value. And some of tomorrow’s biggest winners may be companies we’re barely paying attention to today.
So the next time AI stocks soar or collapse, it’s worth asking yourself two questions. Did the economics of these businesses actually change, or did the market simply change its mind? And how much of what I believe about the future do I actually know?
🌹
Worth reading the original LinkedIn post before piling in. It's a "what if" from someone who works in trading cards. As far as I can tell Nike hasn't announced a card product, and that box looks like a mock-up of where she thinks this could go.
The more interesting bit is .SWOOSH itself. It launched in 2022 as Nike's NFT platform, which most people wrote off. It's now the team that ran the One Piece collab: three Air Max Plus pairs based on Devil Fruits, TCG cards handed out with in-store purchases and matched to the colourway you bought, a Wanted poster printed inside the shoebox, and pop-ups in nearly two dozen cities.
That's cards as a way to sell shoes, which is a very different thing from cards as a new category. Costs almost nothing, gives people a reason to turn up on launch day, and the product stays scarce.
I'd also push back on this being what blows out the brand. The damage over the last few years came from volume. Dunks, AF1s and Jordan 1 Lows everywhere, then sitting in discount bins. Limited drops built around a story work the other way round.
If Nike did go further into cards, the obstacle is licensing. Fanatics/Topps hold the NBA, NFL and MLB card rights, so no team logos. Nike's edge would be its athlete contracts, or more likely a co-brand with Fanatics, who already make Nike's MLB uniforms.
None of this moves the P&L on a mid-$40bn revenue base. Matt's right that the fix is product: running, innovation, getting the big franchises healthy again. I just don't think a box of cards, real or not, is what's standing in the way.
Why can't $NKE just go back to making great product? Why are they so stupidly focused on expanding into every possible category just to blow the brand out even more than it already has been blown out? What is next, a Taco Bell collaboration?
Most cyclical work is triangulation: suppliers on lead times, customers on destocking, competitors on capacity, all layered on top of PMIs, pricing and trade data. That's broad, repetitive cross-referencing, which is what agents are good at. They don't need to call the cycle. They only need to flag a tightening balance a quarter before it shows up in consensus.
But the money in cyclicals usually isn't made by spotting the data first. It's made by acting on it early, against consensus, sizing it properly, and holding through two quarters of looking wrong. Sizing is arguably where machines beat humans too, since we're notoriously bad at it.
There's also a second-order effect. If everyone's agents read the same transcripts, the 'early' signal gets priced faster and the edge decays. What's left is proprietary checks, niche physical-market data, and behaviour.
This is the problem I’ve been trying to build around. Rather than one agent hunting for answers, the idea is an interconnected system. Every transcript, filing, trade release and price series gets ingested and linked across companies, suppliers, customers and end markets. That way a throwaway comment from a customer in one industry surfaces as a flag on a supplier three steps up the chain. It's a living map of the cycle rather than a stack of separate models.
The system reads, links and flags on its own, and rules-based sizing keeps the risk honest.
My guess is cyclicals end up less insulated on detection and more insulated on action. Agents compress the time a tightening stays mispriced. The residual edge is having the stomach to be early.
Have been thinking about AI in institutional investing processes and growing use of agents etc. At some point in my career touched every sector (to a degree) outside of fins and HC. Ironically feels like the most insulated space to AI-driven trading is cyclicals and proper supply/demand cycles. Yes, they’ll catch the EPS revisions up and down but harder to see how they take strong views on harder to forecast set of outcomes.
Have been thinking about AI in institutional investing processes and growing use of agents etc. At some point in my career touched every sector (to a degree) outside of fins and HC. Ironically feels like the most insulated space to AI-driven trading is cyclicals and proper supply/demand cycles. Yes, they’ll catch the EPS revisions up and down but harder to see how they take strong views on harder to forecast set of outcomes.