Latest / Deep Values: Value Investing Deep Dives / Chamath Palihapitiya: The AI Private Credit Bubble Explained
Transcript
- 0:00Tech giants like Amazon, Alphabet, Microsoft, I mean they
- 0:03are sitting on some of the largest piles of cash in human
- 0:06history. Right, Absolutely.
- 0:07Massive reserves. Yeah, so why are they suddenly
- 0:09relying on like massive off balance sheet shadow debt just
- 0:13to buy their computer chips? That is the big question right
- 0:16now. Because if you've been watching
- 0:17the whole explosion of artificial intelligence over the
- 0:20last couple of years, you know, building these models is, well,
- 0:23it's wildly expensive. Oh, incredibly.
- 0:26Yeah, and trillions of dollars are just moving behind the
- 0:28scenes to fund this infrastructure.
- 0:30Which really brings us to what I think is the defining financial
- 0:32question of this decade. Yeah, yeah.
- 0:34Are we witnessing A fundamental necessary restructuring of how
- 0:37the world finances cutting edge tech?
- 0:40Or this is the scary part? Is this multi trillion dollar
- 0:43system just an unsustainable House of Cards?
- 0:45A giant bubble. Exactly.
- 0:46A massive credit bubble. Well, that is exactly the
- 0:49mystery we are tackling. In today's Deep Dive, we're
- 0:51going to unpack an October 2026 research report from Social
- 0:55Capital. Right, the one authored by
- 0:56Chamath Palihapitiya. Yes, and the title really just
- 1:00lays it all out. It's called deep dive.
- 1:03Is the AI build out creating a private credit bubble?
- 1:05It's a fantastic paper. It really is.
- 1:08And before we jump into the historical data and some pretty
- 1:11shocking real world case studies, I actually want to
- 1:13mention something relevant for you listening.
- 1:15Oh yeah, yeah, because we're talking about bubbles today,
- 1:17right? If you're an intelligent
- 1:19investor trying to figure out if the broader market is
- 1:21overpriced, you should really check out the Bubble Watch app
- 1:24for Android. Oh, right, Or the Buffett
- 1:26Indicator app on iOS. Exactly.
- 1:28They are great gauges for market pricing, and there's also the
- 1:31Deep Values iOS app for finding the intrinsic value of any
- 1:34stock. Yes, super helpful tools for
- 1:36navigating all this. Definitely we'll have links to
- 1:38all of that plus the Deep Value Intelligence multi agent AI
- 1:42analyst at Deep Values dot AI and our YouTube channel in the
- 1:45show notes. Remember though, this is all
- 1:47just educational, not investment advice.
- 1:49Always do your own due diligence.
- 1:51Always. OK, so back to the massive AI
- 1:55credit bubble. To really grasp the scale of
- 1:58what's happening, you have to understand the staggering volume
- 2:00of money moving right now. Right, moving completely out of
- 2:03traditional commercial banks. Yeah, and into private credit
- 2:07specifically for technology? You really have to look at the
- 2:09Bank for International Settlements data for this, the
- 2:12BIS. OK, what do their numbers look
- 2:14like? Well, back in 2010, the entire
- 2:17United States direct lending market was it was roughly $96
- 2:21billion. OK.
- 2:23By 2025 and into 2026, that market had exploded to $2.5
- 2:27trillion. Wow, wait 96 billion to 2.5
- 2:33trillion? Trillion with AT.
- 2:35That isn't just growth. I mean, that is a completely
- 2:37different universe of capital. It's staggering, and the
- 2:41composition of that pie is shifting just as dramatically.
- 2:44How so? So the slice dedicated strictly
- 2:47to technology and software that surged from about 22 billion in
- 2:512010 to over a trillion dollars today.
- 2:55Over a trillion just for tech. Yeah, we are at a point where
- 2:59nearly 1/4 of all private credit globally is entirely dependent
- 3:02on tech related cash flows. But the money isn't just like
- 3:06spreading out slowly across thousands of tiny little
- 3:08startups anymore, right? No, not at all.
- 3:11We are seeing these massive, unprecedented, singular deals.
- 3:14Yeah. The report highlights the second
- 3:16quarter of 2026, the anthropic transaction.
- 3:19Oh. That deal was wild.
- 3:20It really was. So Apollo Global Management and
- 3:23Blackstone basically orchestrated this $35 billion
- 3:27special purpose vehicle. Right, an SPV.
- 3:29Yeah, and its sole purpose was just to finance Google Tensor
- 3:33processing units, those specialized TPU chips.
- 3:36Exactly to be leased to the AI model developer Anthropic.
- 3:39UBS actually designated that as the single largest private
- 3:42credit transaction in financial history, which is.
- 3:45Just mind bending when you put it in context.
- 3:47Totally. I mean that single $35 billion
- 3:50deal for Anthropics Hardware that was larger than the average
- 3:54quarterly volume for all private equity backed middle market
- 3:57lending combined. Right, one AI hardware deal
- 4:00swallowed more debt than the entire traditional middle market
- 4:03usually sees in like 3 months. It's insane.
- 4:06It illustrates A profound distortion.
- 4:08I mean, this is a market originally designed to provide
- 4:10loans to traditional steady cash generating businesses.
- 4:14Like regional manufacturing or, you know, healthcare chains.
- 4:17Exactly. And it is morphed into the
- 4:19balance sheet provider for the most expensive technological
- 4:22infrastructure in human history. Which OK, brings me back to the
- 4:26question I started with, because the mechanics here still seem a
- 4:29bit counterintuitive to me. How so?
- 4:32Well, you have these hyperscalers, Alphabet, Amazon,
- 4:36Meta, Oracle. Right, the biggest companies in
- 4:38the world. Yeah, they generate massive
- 4:40revenues. Why do they need external
- 4:43private equity firms to build a $35 billion SPV just to buy
- 4:47chips? It's a great question.
- 4:50Why not just buy them outright? They have the cash.
- 4:52It comes down to the sheer velocity of their capital
- 4:54expenditures or you know, their CapEx, OK, the consensus
- 4:58forecast from places like S&P Global, Morgan Stanley, Goldman
- 5:01Sachs, the. Big players.
- 5:02Yeah. They project that aggregate
- 5:04CapEx across these hyperscalers will hit roughly $870 billion in
- 5:092026. Wow.
- 5:11And then accelerate past $1.3 trillion in 2027.
- 5:16Over a trillion dollars in a single year just on
- 5:19infrastructure. Exactly.
- 5:21And if a hyperscaler tries to put that level of spending
- 5:23directly on its own balance sheet, the financial mechanics
- 5:27just breakdown entirely. Because it's just too much money
- 5:30at once. Right.
- 5:31I mean, if they drain their cash reserves to buy the hardware
- 5:33outright, they lose all their liquidity.
- 5:35Right. They need that cash for
- 5:36operations. Exactly, and if they issue
- 5:38standard corporate bonds to finance it, their debt to equity
- 5:41ratio shoots up. Oh, and that makes traditional
- 5:44bondholders nervous. Very nervous, which leads rating
- 5:47agencies to downgrade their corporate credit score.
- 5:50And a credit downgrade is a disaster, right?
- 5:52Because it means borrowing money for literally everything across
- 5:55their entire global business suddenly gets much more
- 5:57expensive. Precisely so.
- 5:59To avoid that massive penalty, they use a special purpose
- 6:03vehicle. The SPV, right?
- 6:05It's a legally distinct shell company.
- 6:07The SPV actually buys the physical silicon and the data
- 6:10center components. OK.
- 6:12Then the tech company Anthropic in our example, they enter into
- 6:15a master equipment lease with that SPV.
- 6:17Got it. So they make regular lease
- 6:20payments. Exactly, and those payments are
- 6:22passed through to pay the principal and interest to the
- 6:25private credit investors who funded the SPV in the first
- 6:28place. OK, that makes total sense.
- 6:30It keeps the debt off the hyperscalers main books,
- 6:33protects their credit rating, but still gets them the hardware
- 6:36they desperately need. Exactly right.
- 6:39But the Social Capital report suggests this specific
- 6:42structural workaround is actually where the cracks are
- 6:45starting to form. Oh, absolutely.
- 6:47And it's. Specifically because of a
- 6:48concept called duration mismatch.
- 6:51Duration mismatch is the core vulnerability here.
- 6:53Can you break that down for us? Sure.
- 6:55Think about traditional project finance like building a
- 6:58municipal power plant. OK, A lender will typically
- 7:01provide 30 year amortizing debt. That means the borrower pays
- 7:05down the principal and the interest steadily over 30 years.
- 7:09Kind of like a standard home mortgage.
- 7:10Exactly like that, but crucially that debt is protected by a 30
- 7:15year non cancellable contract with a municipal government to
- 7:18actually buy the power. So the debt duration perfectly
- 7:22matches the revenue duration. Exactly.
- 7:24The city is legally bounded by power for the exact amount of
- 7:27time it takes to pay off the loan to build the plant.
- 7:29OK, but how does the AI build out handle this because I'm
- 7:32guessing it's not a 30 year match?
- 7:33Not even close. The report analyzes specialized
- 7:36cloud providers. Companies like Core Weave for
- 7:39example, right lenders are extending 5 year debt facilities
- 7:42to these companies to buy AI hardware.
- 7:45OK, five years. But the customer commitments,
- 7:47the actual contracts from developers renting that compute
- 7:50power, they only average about three years.
- 7:52Wait, what? Yeah.
- 7:54So the loan lasts five years, but the guaranteed income to pay
- 7:58for that loan ends after three. That's the Durisha mismatch.
- 8:01You're basically taking out a five year mortgage on a
- 8:03smartphone. Yeah, that's a great way to put
- 8:05it. You are banking entirely on the
- 8:07assumption that you can still rent that smartphone out to
- 8:09someone at a premium in year 4 and five to cover the rest of
- 8:12your payments. But we all know a smartphone is
- 8:14completely obsolete after a few years.
- 8:16Exactly. And that's exactly what happens
- 8:18with this AI hardware, right? I mean, the collateral backing
- 8:21these massive loans isn't a long lasting concrete power plant.
- 8:25No, it's physical hardware GPU's and custom chips, and they
- 8:29depreciate incredibly fast. Right, because denser, faster
- 8:32chips are constantly hitting the.
- 8:33Market Exactly. Historically, when a new
- 8:36generation of chips arrives, the secondary market utilization and
- 8:39the spot pricing for the older chips completely collapse.
- 8:43So the lenders aren't just betting that the AI models will
- 8:46be successful. No, they are underwriting what
- 8:48we call commercial recontracting risk.
- 8:51Meaning they are just praying that raw compute demand stays so
- 8:54outrageously high that in years four and five developers will
- 8:57still be desperate enough to pay top dollar for last generations
- 9:00hardware. Right, just so the borrower can
- 9:02avoid defaulting on the remaining deck.
- 9:05It is a highly speculative bet. Highly you are leasing rapidly
- 9:09depreciating assets on short term contracts using long term
- 9:13debt. And because of this fundamental
- 9:15misalignment, the report isolates 3 flashing warning
- 9:18signs in the market right now, signs which suggest lenders are
- 9:21largely ignoring these risks. Let's walk through those
- 9:24warnings, starting with how this risk is actually being priced.
- 9:27OK, let's do it. Because if you're taking on that
- 9:30kind of highly speculative mismatched debt, you'd expect
- 9:33lenders to charge a massive premium, right?
- 9:35That's what standard finance theory would tell you, yes.
- 9:37A much higher interest rate to cover the absolute danger of the
- 9:40collateral becoming obsolete. You would expect that, but the
- 9:43BIS data from early 2026 shows something entirely different.
- 9:48What did they find? Private loans extended to AI
- 9:51companies carried an average spread of 6.2 percentage points
- 9:54over the base benchmark rate. OK, 6.2.
- 9:57Non AI corporate borrowers were sitting at an average spread of
- 9:596.1 percentage points. Wait, so 6.2% versus 6.1%?
- 10:07Yes. Let me just clarify for anyone
- 10:10who doesn't track bond spreads daily, that's a difference of 10
- 10:12basis points. A mere 10th of a percent.
- 10:15That is just 110th of 1% difference in the cost of
- 10:18borrowing. It's wild.
- 10:20Private lenders are failing to demand an adequate premium for
- 10:23asset level obsolescence, construction complexity, energy
- 10:26delivery risks, all of it. They're just blinding
- 10:29themselves. They are completely substituting
- 10:32the public hype surrounding AI for rigorous credit
- 10:35underwriting. They see the word AI on the
- 10:37prospectus and just treat a rapidly depreciating server rack
- 10:40as if it has the same risk profile as a 30 year toll
- 10:43bridge. Exactly.
- 10:44But this missing premium leads right into the second massive
- 10:46warning sign, which the report calls the dual disruption
- 10:49paradox. This is a fascinating concept.
- 10:52It really is. Looking at this, it feels like
- 10:54lenders have basically walked themselves into a trap.
- 10:56A very expensive trap. Earlier you mentioned that tech
- 10:58and software make up roughly 25% of the aggregate private credit
- 11:02landscape today. Yes, a huge chunk.
- 11:05And a lot of that is legacy enterprise software, right?
- 11:07Software as a Service or SAS? Exactly.
- 11:10Institutional private equity firms spent years acquiring
- 11:13these SAS businesses using leveraged senior loans, and they
- 11:17underwrote those loans assuming that the high gross margins and
- 11:19the recurring subscription revenues would easily pay down
- 11:22the debt overtime. OK, so if I'm a private credit
- 11:25firm, I've got a massive chunk of My Portfolio relying on these
- 11:28traditional software companies just humming along paying me
- 11:31back, right? But now I'm also lending
- 11:34trillions to build out generative AI infrastructure,
- 11:36correct? If generative AI actually
- 11:38succeeds quickly, doesn't it automate the exact workflows
- 11:42those legacy SAS tools charge monthly subscriptions?
- 11:45For yes, that is the first half of the paradox.
- 11:48Oh wow. If generative AI commoditizes
- 11:51those enterprise workflows faster than anticipated, those
- 11:54legacy software margins deteriorate practically
- 11:56overnight. And that triggers distress and
- 11:58defaults across that massive 25% chunk of their existing private
- 12:01debt portfolios. Exactly.
- 12:03Lenders are funding the very technology designed to destroy
- 12:06the business models of their current borrowers.
- 12:09That's, I mean that's wild, but what if AI takes much longer to
- 12:12actually turn a profit? Well, then you face scenario 2.
- 12:16Enterprise AI monetization develops slowly.
- 12:19OK, If these newly financed AI infrastructure companies take
- 12:22years to become profitable, they simply won't have the operating
- 12:25cash flow to refinance their own maturing debt facilities.
- 12:28Right, when those initial three-year contracts expire.
- 12:30Exactly, and that triggers massive defaults on the new AI
- 12:33loans. So it's a structural trap.
- 12:36A total catch 22. If AI disrupts everything
- 12:39tomorrow, their massive legacy software portfolio burns down.
- 12:42If AI takes a decade to figure out how to be profitable, their
- 12:45new multi trillion dollar infrastructure portfolio burns
- 12:47down. That is the dual disruption
- 12:49paradox. The capital destruction risk
- 12:52exists on both ends of the timeline.
- 12:55Man, and because institutional investors are starting to
- 12:58realize they are caught in this trap, we are seeing the third
- 13:01flashing warning sign, right? Yes, a very real reaction in the
- 13:05market. Because in early 2026, something
- 13:08unprecedented happened with these private debt funds.
- 13:11Right for the first time on record, leading semi liquid
- 13:13private debt funds saw net capital outflows.
- 13:16Meaning they were returning more capital to withdrawing investors
- 13:19and they were raising through new subscriptions.
- 13:22Exactly. People wanted out.
- 13:24And just to unpack why that's so dangerous, these funds promise
- 13:27periodic liquidity to their investors, right?
- 13:30Meaning investors think they can pull their cash out when they
- 13:32want to, but the funds are holding fundamentally illiquid
- 13:36multi year compute loans. Which is a huge problem.
- 13:39Which is a huge problem, yeah. If investors keep asking for
- 13:41their money back, asset managers face a severe liquidity
- 13:44mismatch. They literally don't have the
- 13:46cash on hand. Right.
- 13:48They might be forced to halt new commitments or worse, liquidate
- 13:51the secondary loan portfolios at a steep loss just to generate
- 13:55the cash needed to pay off the withdrawing investors.
- 13:57Which means we are already seeing the secondary debt market
- 14:00start to aggressively price in these risks.
- 14:02Yes, it's happening right now. The report makes it clear this
- 14:04isn't just theoretical modeling anymore.
- 14:07The market is actively splitting into winners and losers based on
- 14:10real world constraints. Really is.
- 14:12Let's look at 2 massive multi billion dollar case studies from
- 14:15the report to see how this plays out in practice.
- 14:17OK, so the divergent here is sharp.
- 14:20Let's start with the debt financing for XA is compute
- 14:23clusters. Right.
- 14:24Elon Musk's company, right? In the secondary market, this
- 14:26specific debt is actually trading at a premium above face
- 14:29value. Meaning investors feel
- 14:31incredibly safe holding it to the point they'll actually pay
- 14:34extra for it. Exactly.
- 14:35And I'm guessing that has everything to do with corporate
- 14:37consolidation because SpaceX acquired XAI, meaning Spacex's
- 14:42broader aerospace balance sheet is essentially Co signing the
- 14:45loan. That's it exactly.
- 14:46It acts as a backstop. Spacex's robust corporate
- 14:49balance sheet absorbs the tenant risk OK, effectively converting
- 14:53what would normally be a highly speculative project debt into a
- 14:57high performing stable credit instrument.
- 14:59So having a giant stable corporate parent makes the debt
- 15:01safe? Yes, but contrast that with a
- 15:03second case study, Project Jupiter.
- 15:06Yes, this was a joint venture between Oracle and Open AI,
- 15:09right? An $18 billion debt facility to
- 15:11build a massive 2.45 GW data center campus in New Mexico.
- 15:16A massive undertaking. And this was backed by what
- 15:19lawyers call hell or high water lease covenants from Oracle.
- 15:22Which is supposed to be the absolute strongest legal
- 15:25guarantee available, right? It means Oracle was
- 15:27contractually obligated to service the lease payments
- 15:30regardless of operational readiness, equipment failure,
- 15:33basically anything. Come hell or high water, you are
- 15:35paying this loan. Exactly.
- 15:37Yet despite an ironclad contract from one of the most established
- 15:40tech companies on earth, the report says this $18 billion
- 15:43debt is heavily distressed. Heavily.
- 15:45It's trading at like 89 to 91 cents on the dollar, and
- 15:48Oracle's credit default swap spreads surged.
- 15:51They absolutely spiked. Let's just translate that last
- 15:54point for you listening. That basically means the cost of
- 15:56buying financial insurance against Oracle failing to pay
- 15:59its debts just skyrocketed in the markets, right?
- 16:01Yes, the market suddenly saw a real risk of Oracle defaulting
- 16:04on this specific project. But wait, if the contract
- 16:08legally bound them to pay no matter what, why did the debt
- 16:11tank? Because the digital revolution
- 16:13hit a very tangible physical wall.
- 16:15What happened? Environmental regulations and
- 16:18local opposition successfully blocked the construction of a
- 16:21critical 17 mile natural gas pipeline.
- 16:23A pipeline. Yeah, and that pipeline was
- 16:27strictly required to power the site's generators.
- 16:30Oh man. Because they physically couldn't
- 16:33get power to the facility, Oracle delivered a formal force
- 16:36majeure notice to the developers.
- 16:38Force majeure, an unforeseeable, unavoidable event that legally
- 16:42frees a party from liability. Precisely by invoking it, Oracle
- 16:46legally delay their rent and carry cost obligations.
- 16:49Wow. And the secondary debt market
- 16:51just panicked immediately. Of course, without that
- 16:54pipeline, there was massive uncertainty over when or even if
- 16:58that $18 billion asset would ever generate a single dollar of
- 17:01cash flow. I just have to pause on the
- 17:03sheer irony here it. Is pretty rich.
- 17:06You have a trillion dollar digital AI revolution, right?
- 17:10The pinnacle of complex human technology, and it gets
- 17:13completely derailed by 17 miles of physical pipe in the dirt.
- 17:18It is the ultimate reality check.
- 17:20It absolutely doesn't matter how bulletproof your financial
- 17:22contract is or how sophisticated your generative AI model is if
- 17:26you physically cannot plug the server into the wall.
- 17:29Investment grade tenant commitments simply cannot
- 17:32neutralize execution risks when basic physical utility access
- 17:36fails. Now, if you were listening to
- 17:38this and keeping track, we're talking about billions in
- 17:41distressed, mispriced debt. We've got forced redemptions
- 17:44from panicked investors. We've got structural market
- 17:47paradoxes where lenders lose whether AI succeeds or fails.
- 17:50It's a lot. And we have massive projects
- 17:52failing over simple utility blockages.
- 17:56It leads to the terrifying question, is this 2008 all over
- 17:59again? Naturally, people are asking.
- 18:00That right. Are we looking at a systemic
- 18:02banking crisis just swapping out subprime mortgages for subprime
- 18:05GPUs? Well, the social capital
- 18:07analysis directly addresses this and their conclusion is
- 18:10definitive, no? OK.
- 18:12Yeah, this is an isolated credit bubble, not a systemic banking
- 18:15contagion. Why?
- 18:17Because today's market has significant structural cushions
- 18:20that simply did not exist in the lead up to 2008.
- 18:23OK. What kind of cushions are we
- 18:24talking about? Why won't this take down the
- 18:26broader economy? Well, first you have third party
- 18:29industrial guarantees. These fundamentally alter
- 18:31recovery expectations. How so?
- 18:34Give me an example. Take that $35 billion Anthropic
- 18:37deal we discussed earlier, right Broadcom, the company that Co
- 18:41designed the TPU chips, they agreed to provide residual value
- 18:44guarantees on the senior debt tranches.
- 18:47Meaning the investors who are first in line to get paid back
- 18:50the senior tranches, they actually have a safety net.
- 18:52Exactly. If the AI model fails and the
- 18:55SPV goes bankrupt, Broadcom backstops the residual value of
- 18:58the physical silicon. Oh.
- 19:00That's smart. Right.
- 19:01It creates A defined floor for liquidity.
- 19:04OK. But the second and arguably much
- 19:07more important factor is the operational leverage profile.
- 19:10The leverage, right? Prior to the 2008 subprime
- 19:14crisis, structured investment vehicles operated with leverage
- 19:17ratios that frequently exceeded 30 to one against those subprime
- 19:21mortgages. 30 to 1, yes. And that was funded heavily by
- 19:24traditional commercial banks. So for everyone actual dollar
- 19:28they had, they were borrowing and betting $30.
- 19:31Exactly. That's terrifying.
- 19:33It was, but today's private credit funds are engineered
- 19:36entirely differently. They maintain fund level
- 19:39leverage ratios between just one to 1:00 and 2:00 to 1:00.
- 19:43Oh wow, so way. Lower, drastically lower.
- 19:46And crucially, this is financed primarily through long term
- 19:48limited partner equity commitments.
- 19:50Which means who exactly is putting up the money?
- 19:53We're talking pension funds, sovereign wealth funds and
- 19:56massive institutional allocators.
- 19:58OK, so it's the difference between a desperate retail
- 20:01trader taking out a massive second mortgage to bet on a
- 20:03stock and a syndicate of multibillion dollar pension
- 20:07funds betting their spare cash. That's a perfect analogy.
- 20:10If the AI bet fails, those pension funds take a massive
- 20:13loss, which certainly hurts their returns.
- 20:15Right, it's bad news for them. Bad news for them, but the
- 20:18global banking system doesn't collapse.
- 20:20Got it. Credit distress and massive
- 20:23project write downs in this AI infrastructure space will
- 20:25decimate the returns for those specific limited partners, but
- 20:29it will not 'cause systemic commercial bank insolvencies.
- 20:32The damage is largely contained within the private credit
- 20:34ecosystem. Precisely.
- 20:36So Synthesizing everything we've looked at in this deep dive,
- 20:39we've tracked this historical shift from equity funding to
- 20:42trillions of dollars in structured shadow debt.
- 20:45Yep, we've uncovered the incredibly dangerous duration
- 20:48mismatch of funding rapidly depreciating chips on short term
- 20:51contracts using long term loans. The smartphone mortgage.
- 20:55Exactly. We mapped out that damned if you
- 20:58do dual disruption paradox where legacy SAS portfolios might burn
- 21:02down to build the AI future. And finally, we've seen why,
- 21:06despite all these flashing red lights, this isolated bubble
- 21:09isn't going to crash your local bank.
- 21:11It's contained, but just because it won't crash the banking
- 21:14system doesn't mean A reckoning isn't coming for the tech
- 21:17sector. Right.
- 21:17When does the bill come due? The report makes it clear that
- 21:21the true macroeconomic stress test is looming just over the
- 21:24horizon in 2027 and 2028. Which is barely a couple of
- 21:28years away. Exactly.
- 21:30That is when the $1.3 trillion annual hyperscaler CapEx bill
- 21:35comes due. That is when entirely new,
- 21:37vastly superior chip architectures will arrive,
- 21:40instantly depreciating the collateral backing all these
- 21:43current loans. Oh wow.
- 21:44And most importantly, that is when those initial three-year
- 21:47compute lease agreements expire. The end of the road.
- 21:51Lenders are going to be forced to mark their collateral to true
- 21:54stark reality. The public narrative exuberance
- 21:57is finally going to collide with hard mathematical reality.
- 22:00Exactly right. And as we wrap up today, I
- 22:02really want to leave you listening with one final thought
- 22:04to Mull over connecting back to the most surprising piece of our
- 22:07discussion, honestly, the pipeline.
- 22:09The pipeline. If tangible physical constraints
- 22:12like a single 17 mile natural gas pipeline can completely
- 22:16derail an $18 billion project backed by some of the most
- 22:19powerful companies on Earth, what happens when the almost
- 22:23unfathomable energy demands of the 2027 AI build out start
- 22:28competing directly with the power grids of everyday cities?
- 22:31The physical infrastructure is the ultimate undeniable
- 22:34bottleneck. Will the ultimate cap on a is
- 22:36explosive growth be financial, or will it simply be electrical?
- 22:40It's definitely something to think about every time you plug
- 22:42in your devices. Absolutely.
- 22:44Thank you so much for joining us as we impact the hidden
- 22:46mechanics of the AI boom. It's been a fascinating shift to
- 22:49analyze. Thanks for having me.