Latest / Deep Values: Value Investing Deep Dives / Howard Marks Warns Against AI Mania
Transcript
- 0:00Howard Marks built a legendary financial career by being, you
- 0:04know, the most patient man in the.
- 0:06Room. Oh, absolutely.
- 0:08He thinks in terms of business cycles.
- 0:10We're talking years, decades even.
- 0:13He's a guy who waits for the dust to settle before he speaks,
- 0:16right? So when a mind that deliberate
- 0:18suddenly rips up his calendar to issue an emergency update just
- 0:2211 weeks after his last one, will you stop what you're doing
- 0:24and listen? It is a genuinely jarring
- 0:27signal. I mean, in the landscape of high
- 0:30level financial observation, patience isn't just a virtue,
- 0:33it's a structural requirement. You wait for quarterly earnings,
- 0:36you observe the capital flows, and then you synthesize.
- 0:40When that rhythm is violently broken like this, it implies the
- 0:44underlying machinery of the world is moving faster than the
- 0:47financial models were even built to track.
- 0:48OK, let's unpack this because that is exactly what brings us
- 0:51to today's deep dive. We are looking at a 46 minute
- 0:54audio release from February 26th, 2020.
- 0:57Sixth. Episode 76.
- 0:59Right, Episode 76 of The Memo by Howard Marks, available globally
- 1:04on Apple Podcasts. It's titled AI Hurdles Ahead,
- 1:08and our mission here is to decode Mark's latest thesis and
- 1:11figure out what his framework means for you as you try to
- 1:13navigate the absolute wall of noise surrounding artificial
- 1:17intelligence. And there's a lot of noise right
- 1:19now. So much noise.
- 1:21But before we get to the tech itself, we have to talk about
- 1:23that timeline. 11 weeks. He published this just 11 weeks
- 1:28after his previous memo, and he explicitly cited massive changes
- 1:32in that tiny window. It's like, well, it's like
- 1:34updating a printed encyclopedia. All the ink is still wet.
- 1:37That's a great way to put it. And what's fascinating here is
- 1:40that in the traditional financial world, 11 weeks is
- 1:43effectively A rounding error, right?
- 1:45It is less than a single fiscal quarter.
- 1:47Corporate strategy simply doesn't pivot in 11 weeks.
- 1:49Yeah, that takes months or years.
- 1:51Exactly. So to step out and declare that
- 1:53the landscape has shifted massively in such a compressed
- 1:55time frame, it tells us that the metrics marks traditionally
- 1:59relies on are lagging way too far behind the reality of the
- 2:02technology. I hear that, but I have to play
- 2:04devil's advocate here for a second.
- 2:06Is it possible that even someone as disciplined as Howard Marks
- 2:10is just getting swept up in the frantic tech news cycle?
- 2:14It's a fair question. Because, you know, you and I
- 2:17both know the echo chamber is deafening right now.
- 2:19Every single day there's a new benchmark, a new model, a new
- 2:23headline screaming about how AI has revolutionized another
- 2:27industry. Oh yeah?
- 2:28Are we actually seeing fundamental massive changes in
- 2:31just 2 1/2 months? Or is the financial sector just
- 2:35doom scrolling like the rest of us?
- 2:36Well, that is the exact trap we have to avoid falling into, and
- 2:39it takes a lot of intellectual honesty to separate the two.
- 2:42It is incredibly easy to mistake a loud software update for a
- 2:47real paradigm shift. But if you look at Marx's
- 2:50rationale, he's looking at the actual essence of the
- 2:53technology's compounding capabilities, not the press
- 2:56releases. The speed of traditional
- 2:58commerce is linear. Right, it takes time to build a
- 3:00factory. Exactly.
- 3:01Or to launch a product line. But the speed of AI development
- 3:05is algorithmic. It's compounding.
- 3:08So what might look like daily noise to a retail observer is,
- 3:12to a rigorous analyst, a compounding curve that is
- 3:15steepening by the week. He's recognizing that the
- 3:18fundamental capacity of these models to process and generate
- 3:21information is iterating faster than standard quarterly
- 3:25financial reports can even measure.
- 3:28So if the technology is compounding that aggressively,
- 3:31what exactly is the nature of the thing that is changing?
- 3:34I mean, he doesn't just wave his hands and say, oh, the tech is
- 3:36getting better. No, not at all.
- 3:37He provides a very specific breakdown of its trait.
- 3:40Right. He isolates its immense power,
- 3:42its speed. And then, and this is the
- 3:44concept that really anchors the whole memo.
- 3:46It's autonomy. Is the keyword.
- 3:48Here, but crucially, he locks those 3 traits directly,
- 3:51alongside a heavy emphasis on its limitations.
- 3:54Yeah, and if we connect this to the bigger picture, binding
- 3:57those concepts together is what prevents this memo from drifting
- 4:00into utopian science fiction. Right.
- 4:02He keeps it grounded. Exactly.
- 4:04He is grounding A profoundly capable technology in stark
- 4:09reality, because autonomy is the dividing line between AI and
- 4:14historical software, but it is also the source of its greatest
- 4:18vulnerabilities. Let's drill down into that
- 4:20autonomy, because the mental models we have for technology
- 4:23don't quite fit here. How do you mean?
- 4:25Well, think about traditional software or even the Internet
- 4:28itself. I look at those like a massive,
- 4:30incredibly complex system of train tracks.
- 4:33OK, I like. You can build incredible routes,
- 4:35you can move massive amounts of data at lightspeed, but the
- 4:39train only goes where the human engineers laid the tracks.
- 4:43Right, it requires explicit coded instructions for every
- 4:46single junction. Exactly what Marx is pointing
- 4:48out with AI's autonomy is that it's no longer a train on a
- 4:51track, it is an off road vehicle mapping its own terrain in real
- 4:56time. That's a huge distinction.
- 4:58Yeah, it doesn't need tracks to navigate unstructured data,
- 5:01which is the source of its immense power.
- 5:03Previous innovations like a calculator, they were like
- 5:06incredibly fast bicycles, but they still needed a human to
- 5:09pedal. And AI's autonomy means the
- 5:11bicycle is starting to steer itself.
- 5:13Exactly. That off road capability
- 5:15fundamentally changes the relationship between user and
- 5:18tool. I mean, historically, human
- 5:21beings have spent millennia mastering tools that amplify our
- 5:24efforts, but those tools were entirely passive.
- 5:27Like a spreadsheet. Right.
- 5:29A spreadsheet only calculates when you input the formula.
- 5:32A search engine only retrieves when you query it.
- 5:35It waits for you, yes. But AI's autonomy means it is
- 5:39probabilistically generating its own connections, predicting
- 5:43logical steps, and synthesizing outputs without continuous human
- 5:47prompting. It isn't just retrieving a file.
- 5:50No, it is inferring intent and generating novel solutions.
- 5:53We if this thing is an autonomous off road vehicle
- 5:56mapping its own logic, why does Mark spend so much time
- 5:59hammering on its limitations? I mean, if it's as powerful as
- 6:02he says, why is he so focused on the flaws?
- 6:04Because the exact mathematical mechanism that gives it autonomy
- 6:08is the same mechanism that creates its limitation.
- 6:10Wait, really? The same mechanism.
- 6:12Yes, large language models and generative networks operate
- 6:16probabilistically. They are basically predicting
- 6:20the most likely next step based on vast training data.
- 6:24OK, when the data is high quality and the context is
- 6:27clear, that probabilistic generation looks like pure
- 6:32magic. I mean, it writes brilliant code
- 6:35or it drafts A flawless legal summary.
- 6:37It's mind blowing when it. Works, but because it doesn't
- 6:40have true grounded comprehension of the physical world, it can
- 6:44just as confidently predict a logical step that is completely
- 6:47hallucinated. So the autonomy and the
- 6:50limitation are born from the exact same architecture.
- 6:52Exactly. It's brilliant because it
- 6:54guesses, and it's brittle because it guesses.
- 6:56Brittle. That's the perfect word for it,
- 6:58because a single logical error compounding at the speed of an
- 7:02autonomous AI model can wipe out massive amounts of value in
- 7:05seconds. Oh wow.
- 7:06Think about it. If a human makes a mistake on a
- 7:08spreadsheet, it's localized. Right.
- 7:10You just fix the cell, yeah? But if an autonomous agent
- 7:13acting on behalf of a financial institution hallucinates a
- 7:16market condition and executes trades based on it.
- 7:19Oh man, the fallout is instantaneous.
- 7:22And catastrophic? Yeah, that brittleness is what
- 7:26is terrifying to a risk averse investor like Marx.
- 7:30So understanding AI according to this framework means holding
- 7:33those two opposing realities in your head at all times.
- 7:36Exactly. It is a completely new breed of
- 7:39autonomous tool, but if you treat it as omnipotent, you will
- 7:43make catastrophic decisions. But if you dismiss it because of
- 7:46its hallucinations, you will be run over by the massive changes
- 7:50it is driving. Right, you have to thread that
- 7:51needle. Here's where it gets really
- 7:53interesting. That tension between an
- 7:55autonomous force that is undeniably changing the world
- 7:58and a brittle tool prone to unprompted errors.
- 8:01It makes my head spin when we start talking about investing.
- 8:04That's a lot to process. Because if this technology is
- 8:07compounding at lightspeed but occasionally hallucinating, how
- 8:10do you even begin to assign a dollar value to the companies
- 8:13building it? It feels like trying to price a
- 8:15hurricane. And this raises an important
- 8:17question, which brings us right to the core of Mark's warning.
- 8:20This is where the philosophy of the value investor meets the
- 8:23frenzy of the market, and it is perhaps the single most
- 8:27difficult cognitive leap for anyone observing A technological
- 8:32revolution. Because his ultimate investment
- 8:34implication is very stark. On one hand, he expresses
- 8:37absolute, unwavering conviction that AI is far from a fad.
- 8:42He plants his flag. He does.
- 8:44It will transform the global economy.
- 8:47But on the exact same page, he issues a massive warning.
- 8:50He says investors must weigh that transformational potential
- 8:54against a strict, rigorous assessment of the
- 8:57appropriateness of the prices of AI assets.
- 8:59It. Requires A fundamental rewiring
- 9:01of how you perceive value. The determining factor of a
- 9:04successful investment is not the quality of the technology.
- 9:07Let me reiterate that because it is the anchor of the entire
- 9:10memo. Go ahead.
- 9:11A great technology does not automatically equal a great
- 9:14investment. I hear you, but man, that's
- 9:16incredibly hard to swell in practice.
- 9:17Oh, definitely. When you check your retirement
- 9:19portfolio and see AI stocks driving the numbers into the
- 9:22stratosphere, or you see a company releasing a model that
- 9:25does the work of 50 junior analysts in three seconds, the
- 9:28natural human instinct is to buy that stock immediately.
- 9:31FOMO is real. Exactly.
- 9:33You want to own a piece of the future.
- 9:36So how does a rational observer separate the reality that AI
- 9:40will change the world from the danger that the company's
- 9:42building it might be vastly overpriced?
- 9:45You untangle it by looking at history, specifically at how
- 9:48markets pricey in expectations. OK, give me an example.
- 9:51Think about the UK railway mania in the 1840s.
- 9:55The railways. Yeah, the railways completely
- 9:58revolutionized human transport, commerce, and communication.
- 10:02They were the AI of their. Day an undeniable world altering
- 10:05technology. Exactly.
- 10:07But the investors who bought into the hype lost absolutely
- 10:10everything. Right.
- 10:11Really. Everything.
- 10:12Yes. Why?
- 10:14Because the price they paid for the railway stocks assumed A
- 10:17flawless future. The price is assumed infinite
- 10:20demand, zero competition and perfect execution.
- 10:24The technology succeeded wildly, but the investors went bankrupt
- 10:28because the capital expenditure was massive and competition
- 10:31drove margins into the floor. And that is exactly what Marx is
- 10:33warning about today with AI. So we are talking about
- 10:36evaluation multiples. If a company is trading at say,
- 10:3950 or 100 times its projected future earnings because everyone
- 10:43assumes AI will give them a monopoly with 0 marginal cost,
- 10:48that price tag requires perfection.
- 10:50You hit the nail on the head. A sky high multiple means all
- 10:54the future success is already priced into the stock today,
- 10:56right? If an AI infrastructure company
- 10:59is priced perfectly, the market is assuming it Will Florida
- 11:02State realize its potential without any hiccups, without
- 11:05regulatory crackdowns, and without a cheaper open source
- 11:09competitor undercutting them next week?
- 11:11But we already established in the previous section that the
- 11:13technology is inherently brittle, it has limitations, it
- 11:17will hallucinate, it will face massive compute costs, and there
- 11:21will obviously be regulatory roadblocks.
- 11:23And if you paid a price that assumed perfection, you suffer
- 11:27the financial consequences of those inevitable bumps in the
- 11:30road. Even if the AI technology
- 11:32ultimately succeeds in the long run, if you buy an asset with no
- 11:36margin of safety, you are taking on all the downside risk of the
- 11:40technology's brittleness while capping your upside because you
- 11:43already paid for the brilliance. That is a staggering realization
- 11:47to sit with the idea that you could be 100% correct about the
- 11:50future of the world, right? You could be totally right that
- 11:53AI is not a fad, that it's an autonomous force that will
- 11:56transform every industry on earth.
- 11:58And you could still lose your shirt because he didn't assess
- 12:01the appropriateness of the prices.
- 12:02It highlights the extreme discipline required to navigate
- 12:07a period of rapid technological change.
- 12:09Yeah, you have to be visionary enough to recognize the immense
- 12:13power and speed of the innovation, but grounded enough
- 12:17to look at the valuation multiples and say no, the math
- 12:20doesn't work at this price. The technology is unprecedented,
- 12:22but the math isn't. Right.
- 12:24The technology is entirely unprecedented.
- 12:26The rules of mathematics, cash flow and valuation are not.
- 12:29I want to push back on that last point though.
- 12:31Are the rules of valuation really immune to this?
- 12:34Let's hear it. Because software is not laying
- 12:37physical steel tracks like the railways did in the 1840's, the
- 12:40margins are different, the distribution is instantaneous,
- 12:44and the capital expenditure, while massive for compute, it
- 12:48scales globally in a way physical infrastructure never
- 12:50could. That's true.
- 12:52So isn't it possible that AI is so unprecedented that it
- 12:55actually justifies these historically insane multiples?
- 12:59It is a fair counter argument, and it is exactly the narrative
- 13:02driving the current market frenzy.
- 13:04The argument is that autonomous AI breaks the traditional
- 13:07economic constraints of Labor and capital, therefore
- 13:10traditional multiples don't apply.
- 13:12Right, the classic this time is different argument.
- 13:14Yes, but Marks as a seasoned distressed debt and value
- 13:18investor has seen the This Time is Different movie play out
- 13:21across multiple decades and multiple asset classes.
- 13:24Oh for sure. Whether it's the Nifty 50 in the
- 13:2670s, the.com bubble in 2000, or the AI boom today, the
- 13:30underlying financial gravity remains.
- 13:32Eventually, reality sets in. The business eventually has to
- 13:36generate cash flow that justifies its enterprise value.
- 13:40AI might alter the speed at which a company reaches scale or
- 13:43the efficiency with which it operates, but it does not repeal
- 13:47the mathematical reality that paying $100 for $1.00 of future
- 13:51earnings is a highly precarious position.
- 13:54Because if anything goes wrong. If the growth decelerates even
- 13:58slightly, or the AI models become commoditized rather than
- 14:01proprietary, the multiple collapses and the investor gets
- 14:05crushed. Which perfectly frames the
- 14:07ultimate take away for everyone listening.
- 14:09So what does this all mean for you?
- 14:11Right. Why does this matter today?
- 14:12Why does a 46 minute audio release from February 2026
- 14:16matter when you check your brokerage account tomorrow or
- 14:19when you sit in a strategy meeting at work?
- 14:21It matters because you can adopt Marx's dual lens framework.
- 14:25Exactly to protect yourself from the hysteria.
- 14:27The dual lens framework is essential for maintaining your
- 14:30equilibrium when the rest of the market is losing its collective
- 14:33mind. Lens number one, you must
- 14:36respect the immense autonomous power of AI, acknowledge its
- 14:40compounding speed, internalize the reality that we are dealing
- 14:43with a tool that steers itself, that infers logic, and that is
- 14:46absolutely not a passing fad. If you ignore it, you will
- 14:50become obsolete. But the second you acknowledge
- 14:52that power, you have to immediately look through lens #2
- 14:56remain highly skeptical, highly disciplined, and completely
- 15:00unsentimental about the priced in expectations the market is
- 15:03assigning to these AI assets. Remember the inherent
- 15:05brittleness of the models. Remember the massive compute
- 15:09costs and the open source competition?
- 15:11Above all, remember that a world altering technology does not
- 15:14automatically guarantee a profitable investment.
- 15:17Your job is to stay anchored to the fundamental appropriateness
- 15:20of the price, even when the technological leaps every 11
- 15:24weeks tempt you to abandon your discipline.
- 15:26It is brilliant, sobering advice from a guy who has survived
- 15:30every major market cycle of the last 50 years.
- 15:33It really is. But as we wrap up this deep
- 15:35dive, I want to leave you the listener, with a lingering
- 15:38thought to Mull over on your own.
- 15:41Oh. All right, we started this
- 15:42conversation by looking at Howard Marks, a man famous for
- 15:46his patients, being forced to RIP up his timeline and issue an
- 15:49update just 11 weeks after his last one.
- 15:52Yeah. Breaking his own rules.
- 15:54The speed of the technology broke his traditional calendar
- 15:57and yet his core advice is to rely on traditional human made
- 16:01metrics of financial valuation to diligently assess the
- 16:05appropriateness of prices. Right, applying the old rules to
- 16:08the new tech. But if artificial intelligence
- 16:11truly possesses the unprecedented autonomy and
- 16:14compounding speed that Marx describes, if it is
- 16:17fundamentally rewriting his own capabilities every few months,
- 16:20how long until those traditional human made metrics for assessing
- 16:23prices stop working entirely? That is a wild thought.
- 16:27If AI continues to hurdle ahead, mapping its own logic and
- 16:31accelerating the global economy at a pace the human brain can
- 16:33easily track, will we eventually reach a point where autonomous
- 16:37AI disrupts and redefines the very concept of financial
- 16:40valuation itself? I guess we'll have to wait and
- 16:42see. It's definitely something to
- 16:43think about the next time you see a headline about an AI
- 16:46breakthrough or the next time you look at the price tag of an
- 16:50AI asset. Thanks for joining us on this
- 16:52deep dive. Keep your eyes open and keep
- 16:54questioning those multiples.