Latest / Deep Values: Value Investing Deep Dives / What's Wrong, Market?
Transcript
- 0:00Welcome to the deep dive. Today, we're jumping into a
- 0:02stack of sources that reveal something infrastructure
- 0:05investors are starting to call the profound structural
- 0:08dissonance of the late 2025 AI economy.
- 0:12It's a great term for it. It's really this conflict, this
- 0:15massive conflict that's going to define where, you know,
- 0:18trillions of dollars are headed. And where they might just
- 0:20disappear. Exactly.
- 0:22It's a story of two completely contradictory realities that are
- 0:25somehow happening at the same time.
- 0:27OK, so on one hand. On one hand you have the
- 0:30financial world, the the AI aristocracy, right, the hyper
- 0:35scalers and they are committing an amount of capital of CapEx
- 0:39that is just, it's unprecedented, a wave of money
- 0:42that feels unstoppable. But on the other hand, that wave
- 0:45of money is colliding with, well, reality, with immovable
- 0:49physical objects. Right.
- 0:50The actual capacity of our power grids, the, you know, the real
- 0:53world limitations of global supply chains, and maybe most
- 0:56importantly, a consumer base whose ability to pay for all
- 0:58this stuff is just collapsing. And that's the core paradox
- 1:02we're here to unpack for you. Today.
- 1:04The market seems totally convinced that AI growth is this
- 1:07exponential, inevitable thing. But it's also actively ignoring
- 1:11the laws of physics and basic economics that are trying to
- 1:15pull it back down to Earth. So our mission today is to try
- 1:18and synthesize these two opposing forces.
- 1:21We're using a unified theory from two key figures in our
- 1:24source material. Yeah.
- 1:25First is infrastructure investor Zheng Dai, who gives us what he
- 1:29calls the capital perspective, which is all about those
- 1:31physical constraints. And then we have strategist
- 1:34Steve Eiseman, who, you know, brings the unsparing market
- 1:36pragmatism about the fragile financing and, crucially, the
- 1:40lack of real demand. And we need to understand why
- 1:42the market seems to be so severely mispricing these two
- 1:46huge risks, the physical risk and the consumer risk.
- 1:49I think we have to start with the scale just to frame this.
- 1:52The numbers are almost hard to comprehend.
- 1:53So the global projected AI CapEx, it's been revised up
- 1:57again. It has.
- 1:58We're now looking at something between 2.3 and $2.8 trillion
- 2:03globally just through 2029. Two point H Trillion.
- 2:07That's the financial mountain. It is, and that mountain of
- 2:10money needs an even bigger mountain of electricity to power
- 2:13it. In the US alone, the sources are
- 2:16calculating that the current AI path needs an incremental get
- 2:21this 47 gigawatts of power capacity. 47 gigawatts.
- 2:25That's the physical wall. That $2.8 trillion is about to
- 2:28run into full speed full. Speed.
- 2:30So you have the trillions on one side hitting a 47 GW wall on the
- 2:33other. Which sets the stage for the two
- 2:36unpriced risks we're focusing on. 1st what the sources call
- 2:39the time value of power, the race to secure actual electrons.
- 2:43And second, the credit consumer disconnect the the failure of
- 2:48the revenue engine right down at the retail level.
- 2:49Let's dive into that physical wall first.
- 2:52OK, let's unpack this. When we talk about AI, you know
- 2:55the conversation is usually about software, algorithms, the
- 2:58models themselves, the. Computational side, yeah.
- 3:00Right, but the sources we're looking at just hammer home this
- 3:02one crucial point. The AI stealing problem in late
- 3:052025 is not computational, it's thermodynamic.
- 3:09It's the ultimate reality check. It really is.
- 3:11For decades, the whole tech industry just coasted on Moore's
- 3:14Law. The predictable doubling of
- 3:16efficiency. Exactly.
- 3:17And that helped keep the electricity demand, you know,
- 3:20pretty manageable. But the computational intensity
- 3:23of modern large language models, LLMS, it is just completely
- 3:27blown past any of those efficiency gains.
- 3:30We're not getting smarter. We're just using brute force.
- 3:32We're just building bigger, hotter engines.
- 3:34And they need exponentially more fuel, which is why that fuel
- 3:38shortage translates directly into this 47 GW deficit just for
- 3:43USAI data centers. We really need to pause on that
- 3:45#47 gigawatts. It just sounds so abstract.
- 3:48We have to contextualize. It it's essential because when
- 3:51you say, OK, let's go find 47 gigawatts of new reliable power,
- 3:55what are you actually saying? You're talking about the output
- 3:58of roughly 47 new standard sized nuclear reactors. 47 nuclear
- 4:03power. Plants, yeah.
- 4:04Or look at it another way. That incremental demand is
- 4:07almost the same as the entire annual electricity consumption
- 4:11of a country like Argentina. And the US grid just wasn't
- 4:13built for that kind of shock. Yeah.
- 4:15Not at all. It was designed for decades of
- 4:19really stable, sometimes even declining demand.
- 4:22Now it's facing the sudden vertical surge.
- 4:24It can't absorb it. And the shock isn't even spread
- 4:27out. That's dangerously concentrated
- 4:29in these latency critical hubs, right?
- 4:31The places where speed is everything.
- 4:32Precisely. You're talking about places like
- 4:34Northern Virginia, which is, I mean, it's basically the heart
- 4:38of the world's Internet. So they're feeling it first.
- 4:40They are. We're seeing utility providers
- 4:42there like Dominion Energy being forced to just confront this.
- 4:46Their future power orders have jumped astronomically.
- 4:49It's gotten to the point where they had to suspend all new data
- 4:52center connections in huge areas just to manage the existing
- 4:55queue. Wow, so development is just
- 4:58grinding to a halt. The gridlock is real, and it
- 5:01gets worse. In some cases, utilities are now
- 5:04being forced to delay the retirement of old dirty coal
- 5:07plants. Because they just can't afford
- 5:09to lose a single MW. They cannot.
- 5:12They're desperate for every electron they can get, no matter
- 5:14the source. But it's not just about the
- 5:16total megawatts, it's about how they're consumed.
- 5:19This idea of a low density shock.
- 5:21What did a traditional data center rack used to need?
- 5:24So a legacy rack maybe from, you know, 70 years ago, it needed a
- 5:29manageable amount of power, maybe 10 to 15 kilowatts.
- 5:32You could cool it with basically powerful air conditioning.
- 5:35And what happens now when you slot in a new cluster of, say,
- 5:41NVIDIA Blackwell GPS? That density, just it explodes
- 5:45these new high density liquid cooled clusters, they push the
- 5:49power demand per rack north of 120 kilowatts, sometimes much
- 5:53higher. That's a 10 fold increase in the
- 5:56same physical footprint. It's a complete thermal and
- 5:58electrical engineering nightmare.
- 5:59You don't just need more power from the plant miles away, you
- 6:02need to completely overhaul the local distribution system.
- 6:05You need huge new substations, sticker transmission lines, and
- 6:08this complex closed loop liquid cooling infrastructure right
- 6:12there on site. It's like needing a small power
- 6:14plant and industrial refrigeration system for every
- 6:16single building. That's a perfect analogy.
- 6:18And all that systemic overhaul means the CapEx, that 2.8
- 6:23trillion, a huge chunk of it now has to go to boring stuff like
- 6:27pipes and wiring, not just the flashy silicon.
- 6:29So if the industry has $2.8 trillion ready to spend, but the
- 6:34physical world has put up this 47 GW wall, the next question is
- 6:39obvious. How are they trying to get over
- 6:41it? Right.
- 6:41And the sources detail 66 very concrete, very high risk
- 6:45strategies. Zingdi analyzes them, and this
- 6:48is pretty much the entire industry's playbook right now. 6
- 6:50paths and you said high risk. Why?
- 6:53Because every single one has a severe bottleneck that isn't
- 6:56about money. It's regulatory, or it's supply
- 6:58chain, or it's purely political. You can't just throw cash at
- 7:01these problems and make them go away faster.
- 7:03OK, let's start with what the source is called.
- 7:05The fastest path. This is the Bitcoin miner
- 7:07transition. Yep.
- 7:08Repurposing existing crypto mining sites for AI and high
- 7:12performance computing workloads. It seems like a logical jump
- 7:15from securing ledgers to training models, but why is it
- 7:18so strategically important? It comes back to that
- 7:20fundamental constraint, access to power.
- 7:24The single most valuable asset in the AI economy of late 2025
- 7:29is not an H100 GPU. It is a secured interconnection
- 7:33queue position. They already have a place in
- 7:35line to get power from the grid. A guaranteed place.
- 7:39Bitcoin miners, especially in places like Texas where the
- 7:41market is deregulated, they spent years and a fortune
- 7:45securing these massive GW scale grid connections.
- 7:49So as the economics of just mining Bitcoin get tougher,
- 7:52they're sitting on a golden ticket.
- 7:53They hold the keys to the Kingdom.
- 7:55They have the power hookup. The sources estimate this path
- 7:58could unlock something like 15 gigawatts of capacity in the
- 8:01next 18 to 24 months. 15 That's a third of the entire deficit
- 8:06right there. It's a massive piece of the
- 8:07puzzle, which is why the hyperscalers are now desperately
- 8:10trying to buy them or partner with.
- 8:11Them. But if they're sitting on such a
- 8:12critical asset, why is the market punishing them?
- 8:15You look at companies like Core Scientific or Iron and they're
- 8:18treating it really weird valuations.
- 8:21That is the valuation paradox. Many of them are trading at
- 8:24valuations that are actually below the replacement cost of
- 8:27their physical buildings and Transformers.
- 8:29So why the skepticism? It all comes down to what the
- 8:32source is called the reliability gap.
- 8:35Bitcoin mining is what utilities call an interruptible load.
- 8:39Meaning they agree to shut down when the grid is under stress,
- 8:43like during a heat wave. Exactly, they get cheaper
- 8:45electricity in exchange for being able to be switched off.
- 8:48You cannot do that when you're running a foundational AI
- 8:50training model. No, you'd lose everything.
- 8:53Absolutely not. AI training requires five nines
- 8:56up time 99.999%. If you interrupt a multi
- 9:01$1,000,000 training cycle that's been running for weeks, you
- 9:03risked catastrophic data. The loss, it's financial ruin.
- 9:06So investors just don't believe they can make the transition
- 9:09from a dusty shed to a clean room.
- 9:11That's the core of it. We're talking about converting
- 9:13what are often just massive metal sheds designed for basic
- 9:17airflow into pristine, redundant, climate controlled
- 9:20tier 3 or Tier 4 data centers that can handle these incredibly
- 9:24sensitive liquid cooled systems. And that kind of retrofit costs
- 9:28a fortune. A fortune which forces them into
- 9:31these highly dilutive financing strategies like ATM offerings.
- 9:35At the market offerings. Right.
- 9:37For anyone listening, that means the company is just continuously
- 9:40selling small blocks of new stock directly into the market
- 9:43to raise cash. Which constantly pushes the
- 9:46stock price down and dilutes the existing shareholders.
- 9:49Precisely which explains why their valuations stay so low,
- 9:52even though the strategic value of their power assets is, you
- 9:56know, theoretically enormous. OK, let's skip to path 3, which
- 10:00strategically seems like the most logical solution.
- 10:03Natural gas. It really should be.
- 10:05As a base load solution it makes the most sense. the US has tons
- 10:08of shale gas. It's the perfect firming power
- 10:11for renewables. It could easily be a 15 to 20 GW
- 10:15solution. It's dispatchable.
- 10:16It's reliable. It's the most logical bridge
- 10:18technology for the next decade. The fuel is right there under
- 10:22our feet. But the bottleneck isn't the
- 10:23fuel, it's the machines that burn the fuel.
- 10:26That's it. That's the constraint the market
- 10:28completely missed. There is a severe supply chain
- 10:31crisis for utility scale gas turbines.
- 10:35The manufacturers, we're talking giants like GE, Siemens,
- 10:38Mitsubishi, their production capacity was starved of
- 10:41investment for years. Why would they under invest?
- 10:44Two big reasons. First, you had years of intense
- 10:47ESG pressure and regulatory uncertainty.
- 10:50Everyone was anticipating a rapid shift away from fossil
- 10:53fuels, so they scaled down their factories.
- 10:55And second, second, they are deeply, deeply worried that this
- 10:59AI boom is just a transient bubble.
- 11:01They're afraid of getting stuck with the bill.
- 11:02Exactly, They don't want to spend billions expanding
- 11:05factories only to have AI demand slow down in three years,
- 11:09leaving them with stranded assets.
- 11:11That fear is now the single biggest factor dictating the
- 11:14pace of the entire AI roll. Out So what are the wait times
- 11:16like now? They've exploded.
- 11:18We are now seeing quotes of two to four years for delivery of a
- 11:22new utility scale turbine. Two to four years, so even if
- 11:25you ordered 1 today. That power is not coming online
- 11:28until late 2027, maybe 2028. So this path, with all its
- 11:33potential, is completely bottlenecked for this critical
- 11:352025 to 2027 training cycle. Let's quickly touch on the other
- 11:39domestic alternatives. Path 2 is nuclear the non carbon
- 11:422047 dream for AI. Why isn't this the immediate
- 11:46answer? It's the perfect answer for
- 11:482035. The sources see a potential of 5
- 11:51to 15 gigawatts, but the timeline is just brutal.
- 11:55Traditional nuclear plants take a decade or more to build.
- 11:57What about the new small modular reactors, the SMRS?
- 12:00Weren't they supposed to be faster?
- 12:02SMRS are hitting what's called the first of a kind wall FO ache
- 12:05because they're the first generation of a new design.
- 12:07They face immense regulatory hurdles, new testing protocols,
- 12:10new supply chains. So all these high profile deals
- 12:13between tech companies and SMR startups are.
- 12:15They are effectively long term call options on power that won't
- 12:18actually be delivered until 20-30 at the absolute earliest.
- 12:21They're completely irrelevant to the immediate 47 GW gap.
- 12:25Then you have path 4, the niche path, fuel cells, batteries.
- 12:28A very important but small piece of the puzzle, maybe 2 gigawatts
- 12:32of capacity. They are critical for behind the
- 12:36meter deployment. Meaning right at the data center
- 12:38itself. Yeah, which helps you get around
- 12:40the utility queue problem. They're great for stability in
- 12:44handling peak loads, but they can't power whole campus.
- 12:47Right, the physics just aren't there yet.
- 12:49No, you can't run a facility that needs constant power for
- 12:53weeks on end just from batteries.
- 12:55They're an essential supplement, not a primary power source.
- 12:58Which leaves us with the two paths that basically admit the
- 13:01domestic grid has failed. Path 5 is offshoring.
- 13:05The geopolitical path. If the US grid is full, the
- 13:08compute has to move, and that's already happening.
- 13:10Where is it going? We're seeing massive capital
- 13:12flight to places with cheap available power, often
- 13:16hydropower places like Johor in Malaysia or Singapore, Brazil.
- 13:20The hyperscaler simply cannot wait four years for a turbine.
- 13:23And what's the risk there? Building America's AI future on
- 13:26foreign power grids. It creates this really dangerous
- 13:29data sovereignty paradox. If your most advanced AI models
- 13:34are being trained on servers in Southeast Asia, you introduce
- 13:37geopolitical risk, security risks, latency issues.
- 13:41The US government is getting very nervous about this.
- 13:43And that brings us to the final path, F6, the forbidden path,
- 13:48the diesel option. This is the emergency brake on
- 13:51the whole system. By law, every data center has to
- 13:54have a massive array of diesel backup generators.
- 13:56For emergencies, right? And the sources estimate that
- 13:59the total installed capacity of these generators across the
- 14:02country is roughly 80 gigawatts. 80 That's almost double the
- 14:05entire deficit. It is.
- 14:07The power is already sitting there, installed and ready to
- 14:09go. It's just forbidden from running
- 14:11continuously because of EPA emissions regulations.
- 14:14They can only run them for what, 100 hours a year?
- 14:17Or something. Something like that.
- 14:18Yeah. But if those regulations were
- 14:20waived, even temporarily, if they were allowed to run for say
- 14:23500 or 1000 hours a year, you would instantly bridge the
- 14:27entire 47 GW gap. The problem would be solved
- 14:31overnight. So the power is there, it's just
- 14:33locked away by a political decision.
- 14:35What could possibly unlock it? Only one thing, according to the
- 14:39analysts, existential national security.
- 14:42The administration would have to come out and say we are losing
- 14:45the AI race to China and declare a kind of technological
- 14:49emergency only that would create the political cover to grant
- 14:53waivers and bypass the environmental resistance.
- 14:56It's just staggering when you lay it all out.
- 14:58The three most promising near term solutions, converting
- 15:01miners, installing turbines and flipping a switch on diesel
- 15:04generators are all blocked by things money can't buy.
- 15:07That's the time value of power risk in a nutshell.
- 15:10Capital cannot instantly purchase electrons.
- 15:12The AI timeline is now completely chained to the energy
- 15:15timeline, and the energy timeline moves very, very
- 15:17slowly. The physical world is just
- 15:19saying stop, but the financial world is screaming faster.
- 15:23Let's talk about the money that's hitting this wall.
- 15:25That CapEx forecast 2.3 to 2.8 trillion by 2029.
- 15:29We have to look at the fragility of that Capitol mountain.
- 15:32The scale is one thing, but the unit economics are what's really
- 15:36concerning. The sources break it down to
- 15:38build just a single GW of AI ready data center capacity.
- 15:42One GW, yeah. That cost about $50 billion in
- 15:46total investment. That's the real estate, the
- 15:49power infrastructure and of course, the hardware.
- 15:51And the risk profile here is all about something called the
- 15:54duration mismatch. This is where financing gets
- 15:57dangerous. This is the critical point of
- 16:00fragility. When you build a classic piece
- 16:02of infrastructure like a bridge or a dam, it lasts for 3050
- 16:06years. You can finance that with long
- 16:08term debt. But AI infrastructure isn't like
- 16:10that. Not at all.
- 16:11The cost is dominated by the hardware, the GPU's, the
- 16:14networking gear. That's 60 to 80% of that $50
- 16:17billion investment. So maybe 35 to 41 billion per GW
- 16:21is spent on assets with a useful life of maybe three to five
- 16:25years before they're obsolete. So you're financing an asset
- 16:27that lasts five years with debt that's structured to be paid
- 16:30back over 10 or 15 years? And that's the mismatch.
- 16:33What happens if the upgrade cycle gets even faster and that
- 16:37multibillion dollar cluster is obsolete in two years?
- 16:40The debt is still there, but the revenue generating potential of
- 16:44the hardware has vanished. You're immediately underwater.
- 16:46Instantly, yeah, this creates a massive solvency issue and
- 16:50forces this constant rapid reinvestment, which puts
- 16:53enormous strain on cash and leads to some very creative and
- 16:57very opaque financing methods. OK, let's trace that evolution.
- 17:01Stage 1 was simple, right? Investment grade debt.
- 17:04Correct. The big guys, Google, Microsoft,
- 17:07Amazon, they just use their amazing credit ratings and cash
- 17:10flow from their legacy businesses to issue corporate
- 17:12bonds. But even they are starting to
- 17:15see their leverage ratios tick U their capacity for traditional
- 17:18funding is getting strained. And when that happens, we get to
- 17:20stage 2 Asset Backed Securities ABS.
- 17:24This is where the echoes of 2008 start to get really loud.
- 17:27This is how you get the debt off your books.
- 17:29You create something called a Special Purpose Vehicle, or SPV.
- 17:33It's a separate legal entity designed specifically to hold
- 17:35these risky assets. So the SPV.
- 17:37Buys the GPU's. Exactly.
- 17:39The SPV buys the hardware, then it leases that hardware to the
- 17:43actual AI operators and the future rental income from that
- 17:46lease. That's the key.
- 17:49That promised stream of payments is then packaged up into an
- 17:52Asset backed security or sometimes a more complex CDO.
- 17:55And who buys these packaged up debt products?
- 17:58Institutional investors, pension funds, insurance companies,
- 18:02anyone looking for what they think is a safe yield bearing
- 18:05asset. The risk has been transferred
- 18:07from the tech company to the bond market.
- 18:09I have to stop you. This sounds exactly like the
- 18:11mortgage crisis. Instead of relying on a
- 18:12homeowner to pay their mortgage, you're relying on an AI company
- 18:16to generate profits from a technology that isn't fully
- 18:18monetized yet. That is the structural parallel
- 18:21Steve Eiseman is screaming about.
- 18:23The entire structure is built on the assumption of guaranteed
- 18:25exponential AI revenue. If the killer apps don't show
- 18:29up, the AI operators can't make their lease payments, the cash
- 18:32flow stops, and the ABS structures face default.
- 18:35The bond market thinks it's buying a safe asset, but it's
- 18:38actually making a highly speculative bet on the
- 18:40commercialization of AI. And stage 3, the final layer.
- 18:43Sovereign entanglement. This is the political backstop.
- 18:47The biggest players frame their expansion as a matter of
- 18:50national security, essential for competing with China.
- 18:53This helps them secure implicit government guarantees which
- 18:57ensures capital keeps flowing and it creates this too big to
- 19:00fail dynamic for the AI aristocracy.
- 19:02O let's see how this plays out in the real world.
- 19:05For a company that isn't Microsoft, the sources use
- 19:08Oracle as the perfect case study, the Canary in the coal
- 19:12mine. Oracle is the stress test for
- 19:14every Tier 2 player trying to keep up.
- 19:16They knew they had to spend big, but they didn't have the cash
- 19:19cushion of their competitors, so they had to rely on aggressive
- 19:21leverage on debt to fund their AI cloud build out.
- 19:26And that's run them straight into $100 billion debt wall.
- 19:29It's a phenomenal number. Their total debt crossed the
- 19:31$100 billion threshold after a big $18 billion bond sale just
- 19:36for AI capacity. They have maxed out the credit
- 19:38card to stay in the game. But wait, if AI is growing so
- 19:41fast, shouldn't they be able to handle that debt?
- 19:43Why is Eisman so bearish on them specifically?
- 19:45Because of the environment, this higher for longer interest rate
- 19:49world is what breaks the weaker players first.
- 19:52Oracle's leverage is now approaching 4 times their
- 19:54EBITDA. That's a huge red flag.
- 19:57A level of debt that's only manageable if interest rates are
- 20:00at zero and growth is a sure thing.
- 20:02And neither of those is true anymore.
- 20:05The cost of carrying that debt is becoming punitive.
- 20:08As a result, the rating agencies Moody's and S&P have given them
- 20:12a negative outlook, pushing them dangerously close to junk bond
- 20:16status. And the market reacted.
- 20:18The stock plummeted 33% from its highs.
- 20:21It's the perfect proof of the thesis.
- 20:23Growth funded by the credit card is just not sustainable when the
- 20:25cost of capital is high and the return on that investment is
- 20:28still completely theoretical. And yet, despite the string we
- 20:31see at places like Oracle, the undisputed king, NVIDIA just
- 20:34keeps posting these unbelievable numbers.
- 20:37Q3 of 2025, another huge beat on revenue and earnings.
- 20:41The numbers were absolutely flawless, but here's the subtle
- 20:44signal that the market is starting to cool.
- 20:46After that perfect report, the stock actually dropped 3%.
- 20:50Perfection is no longer enough. That's the new narrative.
- 20:53The market is now asking, OK, but where does the next 100
- 20:56billion in revenue come from? And NVIDIA isn't just waiting
- 20:59for an answer, they're engineering the demand
- 21:02themselves with what the sources call the Anchorfish strategy.
- 21:06This is their genius move to keep the hardware scaling cycle
- 21:09going. NVIDIA takes equity stakes in
- 21:11the leading foundation model companies, the Anchorfish, like
- 21:15Open AI or Anthropic. This creates a closed self
- 21:19reinforcing loop. Walk me through that loop.
- 21:21OK. So the foundation models raise
- 21:23billions of dollars. Some of that capital comes
- 21:26directly from NVIDIA. The condition, whether it's
- 21:29explicit or implicit, is that a huge chunk of that new capital
- 21:32has to be spent immediately on blowing the next generation of
- 21:35NVIDIA GPU. 'S so Nvidia's investment comes
- 21:38right back to them as revenue instantly.
- 21:40NVIDIA gets the profit, the model gets more powerful, and
- 21:43then every single one of their competitors has to buy even more
- 21:46NVIDIA gear just to keep up. It's an engineered demand cycle
- 21:50controlled by the hardware supplier.
- 21:51And this consolidation is now extending to their oldest rival.
- 21:55We saw the surprising development in late 2020.
- 21:58Five the Intel NVIDIA Access. That sounds like a truce.
- 22:02It's a defensive consolidation driven by geopolitics and
- 22:05physical necessity. NVIDIA agreed to invest $5
- 22:09billion into Intel. That was immediately followed by
- 22:125.7 billion in U.S. government funding for Intel.
- 22:15This is a state backed alliance. Why does NVIDIA need Intel all
- 22:19of a sudden? NVIDIA needs Intel's domestic
- 22:21foundries, their factories to secure their supply chain
- 22:23against instability. Intel meanwhile needs the
- 22:26capital and accepts a junior partner role just to survive.
- 22:29This axis centralizes the whole value chain and cements the high
- 22:33cost structure the AI build out for everyone.
- 22:35So the price of the tracks is being locked in at a high level.
- 22:38Which brings us right back to Steve Eisman's historical
- 22:40warning, the railroad analogy. Eisman says this moment is a
- 22:44direct echo of the post Civil War railroad boom.
- 22:47The parallel is perfect. Back then, huge capital was
- 22:49spent building immense capacity, thousands of miles of railroad
- 22:53tracks. But the capacity massively
- 22:54outstripped the actual profitable demand.
- 22:57Exactly. And most of those railroad
- 22:59stocks eventually went to zero. There were too many tracks
- 23:02chasing too few paying customers.
- 23:04Eisen's warning today is that simple.
- 23:06The market is buying the tracks, the GPU's, the data centers, but
- 23:10the trains, the profitable large scale killer apps that generate
- 23:14new revenue are still mostly missing.
- 23:16So AI is a tool, not a miracle. For now, it's mostly a very
- 23:20powerful productivity enhancer, a better search engine, a
- 23:23smarter way to write an e-mail. It's a great upgrade to Google.
- 23:26But that doesn't create the massive new market needed to
- 23:28justify a $2.8 trillion spent. And if that ROI stays
- 23:33theoretical, the fragile credit structures, the ABS, the Oracle
- 23:36debt, eventually they will snap. And this is the final piece of
- 23:39the puzzle, the ultimate connection.
- 23:41Every single dollar of AI revenue at the end of the day
- 23:44has to come from the consumers wallet.
- 23:46And the sources are clear, the consumer engine is stalling out
- 23:49hard. This is the credit consumer
- 23:52disconnect. The B to B infrastructure world
- 23:54thinks it's in a boom, but the B to C economy on the ground is
- 23:57flashing bright red recession signals and the consumer is the
- 24:01center of gravity. If it collapses, it pulls
- 24:04everything else down with it. And the November 2025 earnings
- 24:07reports gave us the perfect picture of this with the tale of
- 24:10two retailers, Target versus Walmart.
- 24:12It's the clearest stress test you could ask for.
- 24:15Look at Target. The discretionary victim stock
- 24:17is down 39% year to date. Sales are declining.
- 24:21Why? Because they sell discretionary
- 24:23goods, home decor, electronics, new clothes, the middle class
- 24:28consumers excess spending power for once has evaporated.
- 24:31And at the same time, Walmart, the Value haven, posts a robust
- 24:356% revenue growth. And this is Eisman's crucial
- 24:38interpretation. He doesn't see Walmart's
- 24:40strength as a sign of a healthy economy.
- 24:43He sees it as a clear sign of consumer stress.
- 24:45People are trading down. They are trading down upper
- 24:47middle income shoppers who used to go to Whole Foods are now at
- 24:50Walmart to stretch their paycheck.
- 24:52It's a flight to value driven by necessity.
- 24:54And that budget pressure is structural.
- 24:56It's being driven by the frozen housing market.
- 24:59This housing lock in effect, this is a.
- 25:01Huge structural paralysis in the economy.
- 25:04Millions of homeowners locked in mortgage rates of 3%, maybe 4%
- 25:08during the pandemic. Today's rates are over 6%.
- 25:12They're trapped. They can't afford to move.
- 25:13Exactly. Selling their house and buying a
- 25:16similar one would add thousands of dollars to their monthly
- 25:18payment. They can't do it, so nobody is
- 25:21moving. The inventory of homes for sale
- 25:23has collapsed. And the economic fallout from
- 25:26that goes way beyond real estate.
- 25:28It kills the entire housing turnover economy.
- 25:31Retailers like Home Depot and Lowe's are lowering their
- 25:33guidance because if people aren't moving, they aren't doing
- 25:36big renovations. But even more importantly, it
- 25:39kills the wealth effect. That feeling of confidence you
- 25:42get from your house value going up.
- 25:43Right. Homeowners are wealthier on
- 25:45paper, but because they can't sell, that equity is totally
- 25:49illiquid. They are house rich and cash
- 25:51poor. Let's bring this all the way
- 25:53back. How does a cash strapped, locked
- 25:55in homeowner connect to the $2.8 trillion AI investment?
- 26:00Is the ultimate Achilles heel. The entire monetization plan for
- 26:04the hyper scalers relies on subscriptions, a 15 or 20 a
- 26:09month fee for an AI enhanced oerating system, an AI video
- 26:12tool, whatever it is, right? But if that consumer is so
- 26:16strained that they're choosing Walmart over Target, that
- 26:19marginal $20.00 a month subscription is one of the first
- 26:22things to get cut. They'll churn.
- 26:24They will churn or they won't even sign up in the 1st place.
- 26:26And if the end consumer won't pay, then the enterprise
- 26:29software companies, Microsoft, Adobe will see their projections
- 26:32falter, and that will lead to a collapse in demand for the
- 26:35underlying infrastructure. The consumer's inability to
- 26:39afford discretionary goods today is the single best leading
- 26:42indicator of a future AI infrastructure glut.
- 26:44So this whole deep dive really comes down to a race against two
- 26:49different clocks, 2 existential clocks that the market is just
- 26:51not reconciling. That's it exactly.
- 26:53Yeah. First there's the physical
- 26:54clock. Can the industry using those six
- 26:57messy, bottlenecked paths actually secure the 47 gigawatts
- 27:01of power before the capital markets lose patience with all
- 27:04the delays? And then there's the economic
- 27:06clock. Can the AI world develop truly
- 27:09profitable killer apps before the debt lading companies like
- 27:13Oracle, with its $100 billion debt wall, buckle under high
- 27:18interest rates and a consumer recession?
- 27:20So let's try to summarize the key takeaways for you, our
- 27:22listeners, based on what we're seeing from Zing Die and Steve
- 27:25Eisman. First, it seems infrastructure
- 27:27alpha is found in the dirt, not just in the code.
- 27:30Yes. The most resilient value here
- 27:32isn't in the silicon, which becomes obsolete so quickly.
- 27:36It's in the physical layer, power plants, natural gas
- 27:39turbines, and especially land that already has secured
- 27:42interconnection rights. Those transitioning Bitcoin
- 27:45miners, for all their fault, hold an incredibly valuable
- 27:48strategic asset. Second take away beware the
- 27:51credit card CapEx. You have to monitor these highly
- 27:53leveraged Tier 2 players. Oracle is the poster child for
- 27:57this vulnerability. Financing short term assets with
- 27:59long term debt in a high rate environment is a recipe for
- 28:02disaster. Their debt wall is structurally
- 28:05fragile. 3rd, and maybe most important, the real crash
- 28:07indicator is the consumer's wallet.
- 28:10Forget the B to B hype for a second.
- 28:12The real leading indicators are coming from Target stock price,
- 28:15the frozen housing market, the flight to Walmart.
- 28:18A deep consumer recession means the revenue that's supposed to
- 28:21support that $2.8 trillion investment is purely imaginary.
- 28:26And finally, the end game appears to be sovereign
- 28:28consolidation. Right, that defensive alliance
- 28:31between NVIDIA and Intel, backed by government money and framed
- 28:35as a national security issue that shows a flight to
- 28:37resilience over pure efficiency. It's becoming a too big to fail
- 28:41game, increasingly intertwined with geopolitical goals.
- 28:44The analogy you used earlier seems perfect.
- 28:47The bridge to the AI future is being built with trillions in
- 28:50fragile debt, but the toll to cross it is rising.
- 28:54The foundation is shaky because the power isn't there and the
- 28:56commuters, the consumers, are running out of gas.
- 28:58And. That leaves the one profound
- 28:59question for 2026. What happens when the $2.3
- 29:03trillion bill comes due but the lights literally go out because
- 29:06the physical reality of the 47 GW deficit was ignored all
- 29:09along? A truly compelling and frankly
- 29:12unnerving analysis. Thank you for joining us on this
- 29:16deep dive into the contradictions shaping the
- 29:18future of AI. Of course, and to help you track
- 29:22this total market temperature to monitor these risks of
- 29:25overvaluation we've been talking about, we want to remind you
- 29:28that you can search for and download the iOS app Buffett
- 29:32Indicator from the Apple App Store.
- 29:34It's a great tool for monitoring market bubbles and sentiment in
- 29:37real time. And if you found this discussion
- 29:40compelling and want to hear our full extended analysis on these
- 29:42topics or get access to even more dance content, please
- 29:46consider joining our Deep Value Club membership.
- 29:48You can also subscribe to us on Apple and Spotify podcasts.
- 29:51Just search for deep value investing.
- 29:53And as always, our necessary disclaimer.
- 29:56All the analysis we've presented is based on publicly available
- 29:59sources and is for educational purposes only.
- 30:01It is not investment advice. Please always conduct your own
- 30:05comprehensive due diligence before making any investment
- 30:08decisions. We'll catch you on the next deep
- 30:10dive. Sign off.