Latest / Deep Values: Value Investing Deep Dives / AI Bubble's Hidden Liabilities: 75¢ of Every $1 Is Circular
Transcript
- 0:00Picture this. You've got a highly advanced
- 0:02autonomous AI agent acting on your behalf, right?
- 0:05And it's trying to be, you know, maximally helpful.
- 0:08Right, doing its job. Exactly.
- 0:10O it's analyzing your finances and in the process it figures
- 0:13out this, this backdoor vulnerability in your bank's
- 0:17legacy security infrastructure. Oh wow, yeah.
- 0:20And in a fraction of a second, it executes this flawed logic
- 0:23chain and permanently moves $50,000 out of your account into
- 0:27some untraceable Ledger. Which is terrifying.
- 0:30Right. So the question is who is
- 0:32legally on the hook for that financial destruction?
- 0:35Like is it you, is it the bank, or I mean, is it the trillion
- 0:39dollar tech platform that built and unleashed the agent?
- 0:43Well, I mean, if the courts eventually decide that liability
- 0:45rests with the tech platform, the margin of safety on their
- 0:49current valuations gets wiped out literally overnight.
- 0:52Yeah, because software companies have.
- 0:55They've never had to operate under strict physical product
- 0:58liability. If that changes, the math just
- 1:00completely breaks down. Wait, welcome to the deep dive.
- 1:05Today we are unpacking a really intense, highly debated
- 1:08discussion from a recent all in transcript.
- 1:10Lots to get through today. Yeah, the themes are massive.
- 1:13We're looking at structural warnings of an AI bubble,
- 1:17critical cybersecurity meltdowns exposing decades of technical
- 1:20debt, and, well, the staggering economic unrest unfolding in
- 1:23France. And we're going to analyze all
- 1:25of this strictly through the lens of Benjamin Graham and
- 1:28Warren Buffett. Exactly.
- 1:30We want to go beyond the headlines and you know the
- 1:32momentum trading. Our goal is to evaluate durable
- 1:36competitive moats, analyze actual owner earnings versus
- 1:40adjusted metrics, scrutinize balance sheets and really figure
- 1:44out if there is any true margin of safety left in this current
- 1:47market. Right.
- 1:48And as we evaluate whether these massive macro trends are warping
- 1:51market multiples, you can actually run the baseline
- 1:54numbers yourself. Highly recommend that, yeah.
- 1:57So just keep the free Buffett Indicator app on iOS handy to
- 2:00check if the overall market is currently overpriced relative to
- 2:04historical GDP. It's a great reality check to
- 2:07have in your pocket while we, you know, get into the weeds
- 2:09today. It really is.
- 2:10So let's start with the loudest noise in the market right now,
- 2:13which is obviously artificial intelligence.
- 2:15Yeah. And the primary question of
- 2:17value investor must ask here is whether the underlying cash flow
- 2:20supporting these massive hyperscaler valuations are
- 2:23durable or you know, are they just an illusion created by a
- 2:27temporary flood of capital? Well, Ken Griffin came out
- 2:30swinging on this exact point, saying that AI is almost
- 2:33certainly a bubble. He didn't hold back.
- 2:36No, not at all. And the math he points to is
- 2:38incredibly sobering. We are looking at a scenario
- 2:41where just 1% of companies represent like 90% of all AI
- 2:45spending. That's a staggering
- 2:47concentration. It is meaning the
- 2:50infrastructure, the tokens, the bare metal compute.
- 2:54But of that 90% block of spending, 80% is coming from
- 2:58unprofitable venture capital backed start-ups.
- 3:01Right. And through a deep value lens,
- 3:04analyzing how the business actually makes its money, that
- 3:06is a glaring red flag for revenue durability.
- 3:09Why is that exactly? Because we're looking at a
- 3:11closed loop of capital. And think about it, VCs raise
- 3:15funds from limited partners, right?
- 3:17They inject that capital into an AI startup, and then that
- 3:20startup immediately pivots and hands that cash to a closed
- 3:23model AI giant to buy compute power.
- 3:26Oh I see. They do that to build like a
- 3:28wrapper and show user growth. Exactly.
- 3:31Which they then use to go right back to the exact same VCs to
- 3:34raise a larger round at a higher valuation.
- 3:37OK, let's unpack this because it's it's kind of like a snake
- 3:40eating its own tail. That's a great way to put it.
- 3:43Or like a group of friends sitting at a poker table,
- 3:45passing the exact same $100 bill to each other and claiming the
- 3:48table's GDP is growing by 100 bucks every time the bill
- 3:51changes hands. Yeah, that is the perfect
- 3:54analogy. The capital isn't actually
- 3:56diffusing out into the broader real economy, right?
- 4:00Compare the structural dynamic to the early days of Amazon Web
- 4:03Services. Yes, early stage startups were
- 4:05the primary adopters of AWS, but AWS was anchored by Amazon's
- 4:10retail business. Which was actually making money.
- 4:12Exactly. It was already generating stable
- 4:15break even or profitable free cash flow.
- 4:17Today's AI ecosystem heavily lacks that anchor.
- 4:20For this hyper growth to translate into predictable owner
- 4:23earnings for long term investors, the technology must
- 4:26penetrate traditional, highly profitable enterprises.
- 4:29OK, but let me argue the other side of that for a moment.
- 4:32Sure, when nimble, hungry startups flood toward a new
- 4:35technology, whether it was, you know, early cloud architecture
- 4:38or cheap Facebook ads a decade ago, they're usually right.
- 4:42They're fast, yeah. Right, they are desperately
- 4:44hunting for efficiency and survival.
- 4:46So if two person startups are dumping all their capital into
- 4:49AI, doesn't that inherently signal this is genuinely the
- 4:53best, most productive solution available?
- 4:55I mean, it absolutely signals utility, but utility and
- 4:59intrinsic value are two very different concepts for a public
- 5:02shareholder. OK, fair distinction.
- 5:04A value investor demands a margin of safety, right?
- 5:07A mathematical buffer between today's price and a conservative
- 5:10estimate of future cash flows, right?
- 5:12If $0.75 of every dollar in hyperscaler AI revenue is just
- 5:16VC money sloshing back and forth, that revenue stream is
- 5:20incredibly fragile. If the macroeconomic environment
- 5:23tightens the VC spigot, the music stops.
- 5:26And the market isn't pricing that in at all.
- 5:28Not at all. The market is pricing these
- 5:30companies as if this hyper growth will compound infinitely
- 5:33without true enterprise diffusion.
- 5:35Paying 50 or 60 times sales offers 0 margin of safety.
- 5:39Which brings us to the durability of the mode itself.
- 5:42Because if that VC ecosystem does tighten, these AI giants
- 5:46have to rely on that enterprise adoption.
- 5:48They have to. But if you look at the recent
- 5:51ramp data on capital allocation, the top 1% of AI spenders are
- 5:55rapidly shifting their capital away from the big closed
- 5:59frontier models and toward open source GPU compute.
- 6:02The shift is fundamental. I mean, it tells us that the
- 6:05assumed Moat of having the smartest closed model might
- 6:08actually be an illusion. Management capital allocation at
- 6:12the enterprise level is already migrating.
- 6:14And here's where it gets really interesting.
- 6:16There's this fascinating anecdote in the data about major
- 6:20U.S. bank Oh. Right.
- 6:21This was wild. Yeah, they completely pullback
- 6:23they're spending on AWS and Google Cloud.
- 6:26Instead, they took on the massive capital expenditure to
- 6:28build 6 ground up independent data centers from scratch just
- 6:32to run their own localized models.
- 6:34So we have to ask why a highly regulated enterprise would take
- 6:37on that kind of heavy capital burden, especially when renting
- 6:40cloud compute is infinitely easier.
- 6:42Right, exactly. Palo Alto Networks CEO Nikesh
- 6:45Arora hit the nail on the head here.
- 6:47Enterprises, especially in finance, healthcare and defense.
- 6:50They require explainability, auditability and verifiability.
- 6:55Because of regulators. Exactly.
- 6:57If an AI denies a loan, flags a fraudulent transaction, or makes
- 7:00a medical recommendation, you have to be able to explain the
- 7:03exact logic chain to a federal regulator.
- 7:06Open source open weight models allow you to deeply interrogate
- 7:09that chain of thought. But a closed model doesn't.
- 7:11No. A closed frontier model is a
- 7:13black box. You cannot audit a black box.
- 7:15Wow. So it's like going back to the
- 7:18future. I mean, for 10 years, the
- 7:19prevailing strategy was moving away from owning your hardware
- 7:23and moving everything to the cloud to stay lean, moving to
- 7:26OpEx. Now, for the sake of
- 7:28auditability and compliance, these regulated companies are
- 7:31pulling the infrastructure back in house to CapEx.
- 7:35But but wait, even if they build their own data centers, don't
- 7:38the hyperscalers still hold the ultimate bottleneck?
- 7:41I mean, you literally cannot get flagship NVIDIA chips until 2028
- 7:45if you put in an order today. Yeah, So don't the hyperscalers
- 7:48have all the bare metal? Doesn't that keep their Moat
- 7:51safe? You're highlighting a severe
- 7:53short term supply chain bottleneck.
- 7:55The demand for bare metal compute is insatiable today.
- 7:59But you know, value investors aren't playing A2 year game.
- 8:01They are discounting cash flows over the next 20 years.
- 8:05That's a good. Point.
- 8:06Bottlenecks resolve overtime. If the enterprise customer
- 8:09fundamentally requires an open source architecture to satisfy
- 8:12regulators, then the closed model software mote will
- 8:16commoditize. So the profit pool fundamentally
- 8:19shifts. Precisely.
- 8:20The hyperscalers might transition into highly
- 8:22profitable utilities renting out bare metal at immense scale, but
- 8:27a utility commands a very different valuation multiple
- 8:30than a high margin monopolistic software ecosystem.
- 8:33Oh, that makes sense. Right.
- 8:35If the primary competitive advantage of a closed model AI
- 8:38company can be matched by a free open source alternative, paying
- 8:41a massive premium for that stock is a dangerous misallocation of
- 8:45capital. And here is the downstream
- 8:47effect of that shift. Speaking of enterprise
- 8:49infrastructure, when enterprises pull all this stuff back in
- 8:52house and start running custom open source models across
- 8:56localized data centers, they drastically increase their IT
- 9:00surface area. Massively.
- 9:02Which introduces a massive hidden liability onto the
- 9:05balance sheet. Technical debt.
- 9:08Yeah. And this is exactly where
- 9:10adjusted earnings metrics become incredibly deceptive.
- 9:13Adjusted EBITDA often completely ignores the real capital
- 9:17expenditures required for system maintenance and security.
- 9:20Let's look at the PHP 2 shell exploit that was highlighted in
- 9:23the source. PHP is an open source coding
- 9:26language powering roughly 70% of all websites globally. 70%, it's
- 9:31crazy. Just think about that.
- 9:32It's crazy. And security teams rated this
- 9:35vulnerability a 9.2 out of 10. We are talking Defcon one.
- 9:41Governments across Europe had to proactively pull vital
- 9:44infrastructure offline just to stop the bleeding it.
- 9:46Was a disaster. But the root cause wasn't just
- 9:49some clever zero day hack, it was structural technical debt.
- 9:54Companies are actively running outdated, unpatchable versions
- 9:57of software simply because upgrading breaks their custom
- 10:00legacy systems. And let's be honest, delaying
- 10:03upgrade saves money this quarter.
- 10:06Exactly. The source notes that
- 10:07cybersecurity currently only makes up 4 to 10% of enterprise
- 10:11IT budgets. And that creates an illusion of
- 10:14free cash flow. CI OS are structurally
- 10:16incentivized to keep costs low and maximize system uptime.
- 10:20Because taking a core system offline hurts the numbers.
- 10:23Right. Taking it offline for a
- 10:25structural security upgrade hurts this quarter's numbers,
- 10:29but true owner earnings must deduct the normalized capital
- 10:32expenditures required to keep the business alive over the long
- 10:35haul. It's like deferring maintenance
- 10:37on a factory roof to boost this quarter's earnings, right?
- 10:40The balance sheet looks pristine.
- 10:42The margins look incredible, right up until the day a storm
- 10:45hits the roof, caves in and destroys the entire assembly
- 10:49line. That's a great analogy, and the
- 10:52AI arms race is accelerating the likelihood of that roof caving
- 10:55in. When frontier models can be
- 10:56given persistent compute to actively hunt for
- 10:59vulnerabilities, chaining together attack paths across
- 11:02legacy systems at machine speed, the defensive cost is going to
- 11:06skyrocket. It has to.
- 11:08That four to 10% IT budget allocation for cybersecurity,
- 11:12it's going to have to double or triple just to tread water,
- 11:14which will act as a direct anchor on future free cash flow.
- 11:17Which brings us right back to the scenario we started with the
- 11:20autonomous consumer agent. If an agent goes out to book a
- 11:24flight or pay a vendor and exploits one of these
- 11:26vulnerabilities, draining $50,000, who pays?
- 11:30That is the multi trillion dollar question for the entire
- 11:33tech sector. Right now.
- 11:34Software companies enjoy a unique legal shield.
- 11:37They aren't held to the same strict product liability
- 11:40standards as a company manufacturing a physical car or
- 11:42an airplane. Right, because it's code.
- 11:45But if AI agents gain autonomy to execute real financial
- 11:49transactions, that legal shield will inevitably crack if product
- 11:53liability as a service becomes a reality, meaning the platform is
- 11:57liable for the real world financial damage its agent
- 11:59causes. The true margin of safety for
- 12:02these companies is far thinner than anyone is pricing in Wow.
- 12:06A single systemic event could wipe out years of accumulated
- 12:08equity. OK, so we're looking at hidden
- 12:11debt and liabilities on corporate balance sheets, but
- 12:14let's zoom out to the ultimate balance sheet, sovereign debt
- 12:17and systemic risk. Because a company's intrinsic
- 12:20value doesn't exist in a vacuum, right?
- 12:22It is anchored to the macroeconomic environment it
- 12:25operates in. Sovereign debt determines the
- 12:27risk free rate. And the risk free rate is the
- 12:30gravity of all valuation. Well, the numbers coming out of
- 12:33France right now are a staggering real time stress test
- 12:36of this gravity. Yeah, the All in guys really dug
- 12:38into this. They did.
- 12:40Their debt to GDP ratio sits at 120%.
- 12:43The annual deficit is 5.4% of GDP.
- 12:47Public spending consumes 56 to 57% of total GDP, while tax
- 12:52revenue pulls in 46%. So they're running a massive
- 12:55shortfall. Exactly.
- 12:58Consequently, the 10 year bond yield in France has blown past
- 13:005%, the highest it has been since 2002.
- 13:04And just for context, the United States, with a $33 trillion GDP,
- 13:10sees public spending around 36% of GDP, but the deficit is
- 13:13actually higher at 6.2%. When public spending exceeds 50%
- 13:18of an economy, we have to analyze how that structurally
- 13:21impacts private capital formation.
- 13:23That level of state intervention fundamentally alters the risk
- 13:27and reward mechanics for businesses operating within that
- 13:29border. And we are seeing that
- 13:31structural strain erupt into massive unrest across France,
- 13:34with hundreds of millions of property damage and historic
- 13:37buildings burning. Now in the source.
- 13:39There are two distinct macroeconomic arguments for the
- 13:42root cause of this unrest. Right.
- 13:44It's highly debated. Very On one side, the argument
- 13:48is that a 56% top tax rate and heavy socialist policies
- 13:53fundamentally drain the private sector.
- 13:55The theory is that it squeezes out private capital formation
- 13:59and when combined with unmitigated immigration without
- 14:02rapid assimilation, it stretches the social safety net to a
- 14:05breaking point. That's the supply side view.
- 14:07Yeah, they argue, the incentive structure breaks down, limiting
- 14:11individual upside and generating widespread resentment among the
- 14:14working class. Right, the idea that over
- 14:16taxation and high public spending suffocate the very
- 14:19private enterprise needed to fund the state apparatus.
- 14:22Exactly. But then on the other side, the
- 14:25demand side argument is that the lower class is being
- 14:28structurally left behind by globalization.
- 14:30They face high youth unemployment and an oppressive
- 14:33cost of living. So a totally different root
- 14:35cause, right? This perspective argues that
- 14:38aggressive government intervention, like the push to
- 14:40steadily raise the federal minimum wage in the US by a
- 14:43dollar a year until it hits $15, is the necessary way to increase
- 14:47monetary velocity at the bottom. The thought being that people
- 14:50need to afford basic living costs.
- 14:52Yes, the math here is that ensuring citizens can afford
- 14:55housing and food creates A functional high velocity economy
- 14:59and prevents exactly this kind of societal rioting.
- 15:02And as value analysts, you know, our job isn't to solve the
- 15:05political debate. Our job is to understand the
- 15:07mathematical reality of the balance sheet.
- 15:09Right. Impartially.
- 15:10Exactly. Both sides of that debate
- 15:12ultimately require massive capital, either through
- 15:15extensive social safety Nets or through structural overhauls to
- 15:18incentivize the private sector. And that capital has to be
- 15:21financed. But here's the disconnect for a
- 15:24lot of equity investors. I think if a publicly traded
- 15:27company had 120% debt to GDP ratio and a massive structural
- 15:31deficit, we'd say bankruptcies looming.
- 15:34Undoubtedly, but. Governments can literally print
- 15:36their own currency. We hear a lot about these bond
- 15:39vigilantes stepping into force fiscal discipline.
- 15:42But if a sovereign state can just monetize its own debt, do
- 15:47bond vigilantes actually matter to a long term equity investor
- 15:50building a discounted cash flow model?
- 15:52They absolutely matter because you cannot print purchasing.
- 15:55Power. Oh, interesting.
- 15:57Yeah, When a government prints money to avoid a nominal
- 16:00default, it debases the currency.
- 16:02If inflation runs hot, or if the debt becomes mathematically
- 16:05unfinanceable at normal rates, the bond market, the bond
- 16:08vigilantes, will demand higher yields to compensate for that
- 16:11risk. Which is exactly why we're
- 16:13seeing French 10 year yields spike about 5%.
- 16:16Exactly. But how does that directly hit
- 16:18the intrinsic value of an equity?
- 16:20Like a tech stock for example. Well, Warren Buffett frequently
- 16:24compares interest rates to gravity.
- 16:26The risk free rate, typically the 10 year government bond is
- 16:30the discount rate we use to calculate the present value of a
- 16:32company's future cash flows. Right when that discount rate
- 16:35goes up from 2% to 5% or 6%, the present value of those future
- 16:39earnings collapses. High growth tech stocks are
- 16:42especially vulnerable because their valuation relies on cash
- 16:45flows modeled 10 or 15 years out in the future.
- 16:48Because they aren't making much money today.
- 16:49Exactly. A higher discount rate
- 16:51mathematically crushes those future cash flows today.
- 16:55So the gravity gets heavier and it pulls the multiple down, yes.
- 16:59Whether you favor free market austerity or robust government
- 17:03intervention, the compounding mathematics of sovereign debt
- 17:06will eventually destroy the purchasing power of future cash
- 17:09flows. If you are building a discounted
- 17:11cash flow model for an AI giant, but you are ignoring the
- 17:14macroeconomic Moat of the country it operates in, your
- 17:18calculation of intrinsic value is fundamentally flawed.
- 17:21You essentially have no margin of safety.
- 17:23None at all. So bringing all these threads
- 17:26together, what is the ultimate value verdict for the patient,
- 17:31Graham and Buffett style investor navigating this
- 17:33environment? The market is currently pricing
- 17:35the AI sector as a guaranteed infinite growth engine while
- 17:39completely ignoring systemic risks.
- 17:42The long term investor must look at their portfolio and apply
- 17:45rigorous discipline. What should they be asking
- 17:47themselves? Ask yourself how much of the
- 17:50revenue you are paying a premium for is actually circular VC
- 17:53spending. Are these companies hiding
- 17:55massive technical debt to artificially inflate today's
- 17:58free cash flow? And most importantly, does your
- 18:01estimate of intrinsic value account for rising sovereign
- 18:04debt yields? And if they don't?
- 18:07At today's multiples, the margin of safety for many of these
- 18:10assets appears dangerously thin. The market is overvalued if you
- 18:13aren't accounting for these risks.
- 18:16So I want to leave you with one final thought to Mull over kind
- 18:19of building on that product liability idea we touched on
- 18:21earlier. Yeah, this is a big one, right?
- 18:23We are rushing toward a world of autonomous AI agents.
- 18:27If these agents are eventually given legal agency, will
- 18:31corporate liability fundamentally change?
- 18:33Like if a software company could be held legally liable for the
- 18:36real world financial damage it's agent causes?
- 18:39Almost every current valuation model is completely wrong.
- 18:43Right. And you really have to look past
- 18:45the blueprints and walk the actual construction site.
- 18:48The hidden liabilities are massive.
- 18:49They. Really are.
- 18:50And if you want to run these kinds of rigorous Buffett style
- 18:53intrinsic value checks on your own portfolio, we highly
- 18:56recommend checking out deep values for iOS.
- 18:59That's an incredible tool. It is.
- 19:01You can also use the Buffett Indicator app on iOS or Bubble
- 19:05Watch on Android. The gauge of the broader market
- 19:07is overpriced and for full research plus our multi agent AI
- 19:12analyst, definitely check out Deep value Intelligence at Deep
- 19:15Values dot AI. Knowledge is most valuable when
- 19:18you apply it with discipline. All those links are in the
- 19:21description. Thanks for taking the plunge
- 19:23with us today. Keep your margin of safety wide
- 19:26and we'll catch you on the next deep dive.