1 00:00:00,120 --> 00:00:03,520 Tech giants like Amazon, Alphabet, Microsoft, I mean they 2 00:00:03,520 --> 00:00:06,000 are sitting on some of the largest piles of cash in human 3 00:00:06,000 --> 00:00:07,120 history. Right, Absolutely. 4 00:00:07,120 --> 00:00:09,800 Massive reserves. Yeah, so why are they suddenly 5 00:00:09,800 --> 00:00:13,960 relying on like massive off balance sheet shadow debt just 6 00:00:13,960 --> 00:00:16,079 to buy their computer chips? That is the big question right 7 00:00:16,079 --> 00:00:17,960 now. Because if you've been watching 8 00:00:17,960 --> 00:00:20,320 the whole explosion of artificial intelligence over the 9 00:00:20,320 --> 00:00:23,760 last couple of years, you know, building these models is, well, 10 00:00:23,760 --> 00:00:25,840 it's wildly expensive. Oh, incredibly. 11 00:00:26,000 --> 00:00:28,000 Yeah, and trillions of dollars are just moving behind the 12 00:00:28,000 --> 00:00:29,640 scenes to fund this infrastructure. 13 00:00:30,040 --> 00:00:32,759 Which really brings us to what I think is the defining financial 14 00:00:32,759 --> 00:00:34,800 question of this decade. Yeah, yeah. 15 00:00:34,800 --> 00:00:37,880 Are we witnessing A fundamental necessary restructuring of how 16 00:00:37,880 --> 00:00:39,480 the world finances cutting edge tech? 17 00:00:40,200 --> 00:00:43,080 Or this is the scary part? Is this multi trillion dollar 18 00:00:43,080 --> 00:00:45,320 system just an unsustainable House of Cards? 19 00:00:45,320 --> 00:00:46,680 A giant bubble. Exactly. 20 00:00:46,680 --> 00:00:49,520 A massive credit bubble. Well, that is exactly the 21 00:00:49,520 --> 00:00:51,880 mystery we are tackling. In today's Deep Dive, we're 22 00:00:51,880 --> 00:00:55,120 going to unpack an October 2026 research report from Social 23 00:00:55,120 --> 00:00:56,720 Capital. Right, the one authored by 24 00:00:56,720 --> 00:01:00,120 Chamath Palihapitiya. Yes, and the title really just 25 00:01:00,120 --> 00:01:02,840 lays it all out. It's called deep dive. 26 00:01:03,200 --> 00:01:05,880 Is the AI build out creating a private credit bubble? 27 00:01:05,960 --> 00:01:08,040 It's a fantastic paper. It really is. 28 00:01:08,320 --> 00:01:11,240 And before we jump into the historical data and some pretty 29 00:01:11,240 --> 00:01:13,800 shocking real world case studies, I actually want to 30 00:01:13,800 --> 00:01:15,480 mention something relevant for you listening. 31 00:01:15,480 --> 00:01:17,880 Oh yeah, yeah, because we're talking about bubbles today, 32 00:01:17,880 --> 00:01:19,560 right? If you're an intelligent 33 00:01:19,560 --> 00:01:21,720 investor trying to figure out if the broader market is 34 00:01:21,720 --> 00:01:24,640 overpriced, you should really check out the Bubble Watch app 35 00:01:24,640 --> 00:01:26,600 for Android. Oh, right, Or the Buffett 36 00:01:26,600 --> 00:01:28,640 Indicator app on iOS. Exactly. 37 00:01:28,920 --> 00:01:31,760 They are great gauges for market pricing, and there's also the 38 00:01:31,760 --> 00:01:34,960 Deep Values iOS app for finding the intrinsic value of any 39 00:01:34,960 --> 00:01:36,680 stock. Yes, super helpful tools for 40 00:01:36,680 --> 00:01:38,880 navigating all this. Definitely we'll have links to 41 00:01:38,880 --> 00:01:42,200 all of that plus the Deep Value Intelligence multi agent AI 42 00:01:42,200 --> 00:01:45,520 analyst at Deep Values dot AI and our YouTube channel in the 43 00:01:45,520 --> 00:01:47,640 show notes. Remember though, this is all 44 00:01:47,640 --> 00:01:49,840 just educational, not investment advice. 45 00:01:49,840 --> 00:01:51,360 Always do your own due diligence. 46 00:01:51,720 --> 00:01:55,280 Always. OK, so back to the massive AI 47 00:01:55,280 --> 00:01:58,080 credit bubble. To really grasp the scale of 48 00:01:58,080 --> 00:02:00,760 what's happening, you have to understand the staggering volume 49 00:02:00,760 --> 00:02:03,920 of money moving right now. Right, moving completely out of 50 00:02:03,920 --> 00:02:07,160 traditional commercial banks. Yeah, and into private credit 51 00:02:07,240 --> 00:02:09,880 specifically for technology? You really have to look at the 52 00:02:09,880 --> 00:02:12,400 Bank for International Settlements data for this, the 53 00:02:12,400 --> 00:02:14,480 BIS. OK, what do their numbers look 54 00:02:14,480 --> 00:02:17,440 like? Well, back in 2010, the entire 55 00:02:17,440 --> 00:02:21,400 United States direct lending market was it was roughly $96 56 00:02:21,400 --> 00:02:23,000 billion. OK. 57 00:02:23,120 --> 00:02:27,920 By 2025 and into 2026, that market had exploded to $2.5 58 00:02:27,920 --> 00:02:33,000 trillion. Wow, wait 96 billion to 2.5 59 00:02:33,000 --> 00:02:34,600 trillion? Trillion with AT. 60 00:02:35,200 --> 00:02:37,880 That isn't just growth. I mean, that is a completely 61 00:02:37,880 --> 00:02:41,200 different universe of capital. It's staggering, and the 62 00:02:41,200 --> 00:02:43,840 composition of that pie is shifting just as dramatically. 63 00:02:44,360 --> 00:02:47,040 How so? So the slice dedicated strictly 64 00:02:47,040 --> 00:02:51,280 to technology and software that surged from about 22 billion in 65 00:02:51,280 --> 00:02:54,240 2010 to over a trillion dollars today. 66 00:02:55,120 --> 00:02:59,080 Over a trillion just for tech. Yeah, we are at a point where 67 00:02:59,080 --> 00:03:02,800 nearly 1/4 of all private credit globally is entirely dependent 68 00:03:02,800 --> 00:03:06,440 on tech related cash flows. But the money isn't just like 69 00:03:06,480 --> 00:03:08,760 spreading out slowly across thousands of tiny little 70 00:03:08,760 --> 00:03:10,880 startups anymore, right? No, not at all. 71 00:03:11,200 --> 00:03:14,360 We are seeing these massive, unprecedented, singular deals. 72 00:03:14,640 --> 00:03:16,040 Yeah. The report highlights the second 73 00:03:16,040 --> 00:03:19,080 quarter of 2026, the anthropic transaction. 74 00:03:19,160 --> 00:03:20,600 Oh. That deal was wild. 75 00:03:20,800 --> 00:03:23,520 It really was. So Apollo Global Management and 76 00:03:23,520 --> 00:03:27,080 Blackstone basically orchestrated this $35 billion 77 00:03:27,400 --> 00:03:29,560 special purpose vehicle. Right, an SPV. 78 00:03:29,680 --> 00:03:33,200 Yeah, and its sole purpose was just to finance Google Tensor 79 00:03:33,200 --> 00:03:36,120 processing units, those specialized TPU chips. 80 00:03:36,520 --> 00:03:39,240 Exactly to be leased to the AI model developer Anthropic. 81 00:03:39,800 --> 00:03:42,840 UBS actually designated that as the single largest private 82 00:03:42,840 --> 00:03:45,280 credit transaction in financial history, which is. 83 00:03:45,280 --> 00:03:47,120 Just mind bending when you put it in context. 84 00:03:47,360 --> 00:03:50,520 Totally. I mean that single $35 billion 85 00:03:50,520 --> 00:03:54,200 deal for Anthropics Hardware that was larger than the average 86 00:03:54,200 --> 00:03:57,360 quarterly volume for all private equity backed middle market 87 00:03:57,360 --> 00:04:00,600 lending combined. Right, one AI hardware deal 88 00:04:00,600 --> 00:04:03,480 swallowed more debt than the entire traditional middle market 89 00:04:03,480 --> 00:04:06,160 usually sees in like 3 months. It's insane. 90 00:04:06,240 --> 00:04:08,280 It illustrates A profound distortion. 91 00:04:08,960 --> 00:04:10,800 I mean, this is a market originally designed to provide 92 00:04:10,800 --> 00:04:14,400 loans to traditional steady cash generating businesses. 93 00:04:14,760 --> 00:04:17,800 Like regional manufacturing or, you know, healthcare chains. 94 00:04:17,920 --> 00:04:19,640 Exactly. And it is morphed into the 95 00:04:19,640 --> 00:04:22,640 balance sheet provider for the most expensive technological 96 00:04:22,640 --> 00:04:26,600 infrastructure in human history. Which OK, brings me back to the 97 00:04:26,600 --> 00:04:29,840 question I started with, because the mechanics here still seem a 98 00:04:29,840 --> 00:04:32,160 bit counterintuitive to me. How so? 99 00:04:32,960 --> 00:04:36,680 Well, you have these hyperscalers, Alphabet, Amazon, 100 00:04:36,680 --> 00:04:38,800 Meta, Oracle. Right, the biggest companies in 101 00:04:38,800 --> 00:04:40,520 the world. Yeah, they generate massive 102 00:04:40,520 --> 00:04:43,320 revenues. Why do they need external 103 00:04:43,320 --> 00:04:47,480 private equity firms to build a $35 billion SPV just to buy 104 00:04:47,480 --> 00:04:49,960 chips? It's a great question. 105 00:04:50,080 --> 00:04:52,040 Why not just buy them outright? They have the cash. 106 00:04:52,720 --> 00:04:54,880 It comes down to the sheer velocity of their capital 107 00:04:54,880 --> 00:04:58,280 expenditures or you know, their CapEx, OK, the consensus 108 00:04:58,280 --> 00:05:01,480 forecast from places like S&P Global, Morgan Stanley, Goldman 109 00:05:01,480 --> 00:05:02,640 Sachs, the. Big players. 110 00:05:02,680 --> 00:05:04,280 Yeah. They project that aggregate 111 00:05:04,280 --> 00:05:09,560 CapEx across these hyperscalers will hit roughly $870 billion in 112 00:05:09,560 --> 00:05:11,040 2026. Wow. 113 00:05:11,600 --> 00:05:16,520 And then accelerate past $1.3 trillion in 2027. 114 00:05:16,640 --> 00:05:19,760 Over a trillion dollars in a single year just on 115 00:05:19,760 --> 00:05:21,000 infrastructure. Exactly. 116 00:05:21,600 --> 00:05:23,880 And if a hyperscaler tries to put that level of spending 117 00:05:23,880 --> 00:05:27,120 directly on its own balance sheet, the financial mechanics 118 00:05:27,120 --> 00:05:30,080 just breakdown entirely. Because it's just too much money 119 00:05:30,080 --> 00:05:31,040 at once. Right. 120 00:05:31,240 --> 00:05:33,480 I mean, if they drain their cash reserves to buy the hardware 121 00:05:33,480 --> 00:05:35,360 outright, they lose all their liquidity. 122 00:05:35,480 --> 00:05:36,440 Right. They need that cash for 123 00:05:36,440 --> 00:05:38,640 operations. Exactly, and if they issue 124 00:05:38,640 --> 00:05:41,640 standard corporate bonds to finance it, their debt to equity 125 00:05:41,640 --> 00:05:44,360 ratio shoots up. Oh, and that makes traditional 126 00:05:44,360 --> 00:05:47,920 bondholders nervous. Very nervous, which leads rating 127 00:05:47,920 --> 00:05:49,880 agencies to downgrade their corporate credit score. 128 00:05:50,480 --> 00:05:52,160 And a credit downgrade is a disaster, right? 129 00:05:52,160 --> 00:05:55,000 Because it means borrowing money for literally everything across 130 00:05:55,000 --> 00:05:57,720 their entire global business suddenly gets much more 131 00:05:57,720 --> 00:05:59,560 expensive. Precisely so. 132 00:05:59,640 --> 00:06:03,040 To avoid that massive penalty, they use a special purpose 133 00:06:03,040 --> 00:06:04,600 vehicle. The SPV, right? 134 00:06:05,240 --> 00:06:06,960 It's a legally distinct shell company. 135 00:06:07,520 --> 00:06:10,400 The SPV actually buys the physical silicon and the data 136 00:06:10,400 --> 00:06:11,640 center components. OK. 137 00:06:12,120 --> 00:06:15,480 Then the tech company Anthropic in our example, they enter into 138 00:06:15,480 --> 00:06:17,640 a master equipment lease with that SPV. 139 00:06:17,880 --> 00:06:20,040 Got it. So they make regular lease 140 00:06:20,040 --> 00:06:22,480 payments. Exactly, and those payments are 141 00:06:22,480 --> 00:06:25,480 passed through to pay the principal and interest to the 142 00:06:25,480 --> 00:06:28,040 private credit investors who funded the SPV in the first 143 00:06:28,040 --> 00:06:30,320 place. OK, that makes total sense. 144 00:06:30,680 --> 00:06:33,240 It keeps the debt off the hyperscalers main books, 145 00:06:33,640 --> 00:06:36,440 protects their credit rating, but still gets them the hardware 146 00:06:36,440 --> 00:06:38,680 they desperately need. Exactly right. 147 00:06:39,600 --> 00:06:42,440 But the Social Capital report suggests this specific 148 00:06:42,440 --> 00:06:45,280 structural workaround is actually where the cracks are 149 00:06:45,280 --> 00:06:46,960 starting to form. Oh, absolutely. 150 00:06:47,200 --> 00:06:48,240 And it's. Specifically because of a 151 00:06:48,240 --> 00:06:50,600 concept called duration mismatch. 152 00:06:51,040 --> 00:06:53,600 Duration mismatch is the core vulnerability here. 153 00:06:53,640 --> 00:06:54,880 Can you break that down for us? Sure. 154 00:06:55,600 --> 00:06:58,840 Think about traditional project finance like building a 155 00:06:58,840 --> 00:07:01,800 municipal power plant. OK, A lender will typically 156 00:07:01,800 --> 00:07:05,720 provide 30 year amortizing debt. That means the borrower pays 157 00:07:05,720 --> 00:07:08,680 down the principal and the interest steadily over 30 years. 158 00:07:09,080 --> 00:07:10,720 Kind of like a standard home mortgage. 159 00:07:10,760 --> 00:07:15,400 Exactly like that, but crucially that debt is protected by a 30 160 00:07:15,400 --> 00:07:18,800 year non cancellable contract with a municipal government to 161 00:07:18,840 --> 00:07:22,440 actually buy the power. So the debt duration perfectly 162 00:07:22,440 --> 00:07:24,320 matches the revenue duration. Exactly. 163 00:07:24,800 --> 00:07:27,280 The city is legally bounded by power for the exact amount of 164 00:07:27,280 --> 00:07:29,720 time it takes to pay off the loan to build the plant. 165 00:07:29,920 --> 00:07:32,240 OK, but how does the AI build out handle this because I'm 166 00:07:32,240 --> 00:07:33,960 guessing it's not a 30 year match? 167 00:07:33,960 --> 00:07:36,840 Not even close. The report analyzes specialized 168 00:07:36,840 --> 00:07:39,120 cloud providers. Companies like Core Weave for 169 00:07:39,120 --> 00:07:42,720 example, right lenders are extending 5 year debt facilities 170 00:07:42,720 --> 00:07:44,480 to these companies to buy AI hardware. 171 00:07:45,000 --> 00:07:47,320 OK, five years. But the customer commitments, 172 00:07:47,320 --> 00:07:50,240 the actual contracts from developers renting that compute 173 00:07:50,240 --> 00:07:52,720 power, they only average about three years. 174 00:07:52,920 --> 00:07:53,960 Wait, what? Yeah. 175 00:07:54,280 --> 00:07:58,440 So the loan lasts five years, but the guaranteed income to pay 176 00:07:58,440 --> 00:08:01,560 for that loan ends after three. That's the Durisha mismatch. 177 00:08:01,840 --> 00:08:03,640 You're basically taking out a five year mortgage on a 178 00:08:03,640 --> 00:08:05,240 smartphone. Yeah, that's a great way to put 179 00:08:05,240 --> 00:08:07,280 it. You are banking entirely on the 180 00:08:07,280 --> 00:08:09,560 assumption that you can still rent that smartphone out to 181 00:08:09,560 --> 00:08:12,720 someone at a premium in year 4 and five to cover the rest of 182 00:08:12,720 --> 00:08:14,880 your payments. But we all know a smartphone is 183 00:08:14,880 --> 00:08:16,800 completely obsolete after a few years. 184 00:08:16,800 --> 00:08:18,960 Exactly. And that's exactly what happens 185 00:08:18,960 --> 00:08:21,840 with this AI hardware, right? I mean, the collateral backing 186 00:08:21,840 --> 00:08:25,120 these massive loans isn't a long lasting concrete power plant. 187 00:08:25,240 --> 00:08:29,200 No, it's physical hardware GPU's and custom chips, and they 188 00:08:29,200 --> 00:08:32,480 depreciate incredibly fast. Right, because denser, faster 189 00:08:32,480 --> 00:08:33,640 chips are constantly hitting the. 190 00:08:33,640 --> 00:08:36,360 Market Exactly. Historically, when a new 191 00:08:36,360 --> 00:08:39,559 generation of chips arrives, the secondary market utilization and 192 00:08:39,559 --> 00:08:43,240 the spot pricing for the older chips completely collapse. 193 00:08:43,880 --> 00:08:46,040 So the lenders aren't just betting that the AI models will 194 00:08:46,040 --> 00:08:48,440 be successful. No, they are underwriting what 195 00:08:48,440 --> 00:08:50,920 we call commercial recontracting risk. 196 00:08:51,160 --> 00:08:54,320 Meaning they are just praying that raw compute demand stays so 197 00:08:54,320 --> 00:08:57,840 outrageously high that in years four and five developers will 198 00:08:57,840 --> 00:09:00,800 still be desperate enough to pay top dollar for last generations 199 00:09:00,800 --> 00:09:02,720 hardware. Right, just so the borrower can 200 00:09:02,720 --> 00:09:04,520 avoid defaulting on the remaining deck. 201 00:09:05,120 --> 00:09:09,800 It is a highly speculative bet. Highly you are leasing rapidly 202 00:09:09,800 --> 00:09:13,360 depreciating assets on short term contracts using long term 203 00:09:13,360 --> 00:09:15,400 debt. And because of this fundamental 204 00:09:15,400 --> 00:09:18,440 misalignment, the report isolates 3 flashing warning 205 00:09:18,440 --> 00:09:21,560 signs in the market right now, signs which suggest lenders are 206 00:09:21,560 --> 00:09:24,280 largely ignoring these risks. Let's walk through those 207 00:09:24,280 --> 00:09:27,480 warnings, starting with how this risk is actually being priced. 208 00:09:27,720 --> 00:09:30,320 OK, let's do it. Because if you're taking on that 209 00:09:30,320 --> 00:09:33,320 kind of highly speculative mismatched debt, you'd expect 210 00:09:33,320 --> 00:09:35,120 lenders to charge a massive premium, right? 211 00:09:35,360 --> 00:09:37,520 That's what standard finance theory would tell you, yes. 212 00:09:37,760 --> 00:09:40,360 A much higher interest rate to cover the absolute danger of the 213 00:09:40,360 --> 00:09:43,800 collateral becoming obsolete. You would expect that, but the 214 00:09:43,800 --> 00:09:48,000 BIS data from early 2026 shows something entirely different. 215 00:09:48,480 --> 00:09:51,080 What did they find? Private loans extended to AI 216 00:09:51,080 --> 00:09:54,320 companies carried an average spread of 6.2 percentage points 217 00:09:54,320 --> 00:09:56,760 over the base benchmark rate. OK, 6.2. 218 00:09:57,000 --> 00:09:59,960 Non AI corporate borrowers were sitting at an average spread of 219 00:09:59,960 --> 00:10:07,320 6.1 percentage points. Wait, so 6.2% versus 6.1%? 220 00:10:07,840 --> 00:10:10,000 Yes. Let me just clarify for anyone 221 00:10:10,000 --> 00:10:12,960 who doesn't track bond spreads daily, that's a difference of 10 222 00:10:12,960 --> 00:10:15,680 basis points. A mere 10th of a percent. 223 00:10:15,840 --> 00:10:18,760 That is just 110th of 1% difference in the cost of 224 00:10:18,760 --> 00:10:20,240 borrowing. It's wild. 225 00:10:20,720 --> 00:10:23,640 Private lenders are failing to demand an adequate premium for 226 00:10:23,640 --> 00:10:26,960 asset level obsolescence, construction complexity, energy 227 00:10:26,960 --> 00:10:29,400 delivery risks, all of it. They're just blinding 228 00:10:29,400 --> 00:10:32,120 themselves. They are completely substituting 229 00:10:32,120 --> 00:10:35,160 the public hype surrounding AI for rigorous credit 230 00:10:35,160 --> 00:10:37,080 underwriting. They see the word AI on the 231 00:10:37,080 --> 00:10:40,400 prospectus and just treat a rapidly depreciating server rack 232 00:10:40,640 --> 00:10:43,320 as if it has the same risk profile as a 30 year toll 233 00:10:43,320 --> 00:10:44,160 bridge. Exactly. 234 00:10:44,400 --> 00:10:46,920 But this missing premium leads right into the second massive 235 00:10:46,920 --> 00:10:49,840 warning sign, which the report calls the dual disruption 236 00:10:49,840 --> 00:10:51,800 paradox. This is a fascinating concept. 237 00:10:52,160 --> 00:10:54,080 It really is. Looking at this, it feels like 238 00:10:54,080 --> 00:10:56,240 lenders have basically walked themselves into a trap. 239 00:10:56,280 --> 00:10:58,960 A very expensive trap. Earlier you mentioned that tech 240 00:10:58,960 --> 00:11:02,840 and software make up roughly 25% of the aggregate private credit 241 00:11:02,840 --> 00:11:04,800 landscape today. Yes, a huge chunk. 242 00:11:05,240 --> 00:11:07,640 And a lot of that is legacy enterprise software, right? 243 00:11:07,640 --> 00:11:10,160 Software as a Service or SAS? Exactly. 244 00:11:10,440 --> 00:11:13,240 Institutional private equity firms spent years acquiring 245 00:11:13,240 --> 00:11:17,120 these SAS businesses using leveraged senior loans, and they 246 00:11:17,120 --> 00:11:19,720 underwrote those loans assuming that the high gross margins and 247 00:11:19,720 --> 00:11:22,560 the recurring subscription revenues would easily pay down 248 00:11:22,560 --> 00:11:25,360 the debt overtime. OK, so if I'm a private credit 249 00:11:25,360 --> 00:11:28,440 firm, I've got a massive chunk of My Portfolio relying on these 250 00:11:28,440 --> 00:11:31,440 traditional software companies just humming along paying me 251 00:11:31,440 --> 00:11:34,000 back, right? But now I'm also lending 252 00:11:34,000 --> 00:11:36,600 trillions to build out generative AI infrastructure, 253 00:11:36,800 --> 00:11:38,960 correct? If generative AI actually 254 00:11:38,960 --> 00:11:42,680 succeeds quickly, doesn't it automate the exact workflows 255 00:11:42,680 --> 00:11:45,440 those legacy SAS tools charge monthly subscriptions? 256 00:11:45,440 --> 00:11:48,480 For yes, that is the first half of the paradox. 257 00:11:48,640 --> 00:11:51,160 Oh wow. If generative AI commoditizes 258 00:11:51,160 --> 00:11:54,240 those enterprise workflows faster than anticipated, those 259 00:11:54,240 --> 00:11:56,760 legacy software margins deteriorate practically 260 00:11:56,760 --> 00:11:58,720 overnight. And that triggers distress and 261 00:11:58,720 --> 00:12:01,800 defaults across that massive 25% chunk of their existing private 262 00:12:01,800 --> 00:12:03,080 debt portfolios. Exactly. 263 00:12:03,640 --> 00:12:06,760 Lenders are funding the very technology designed to destroy 264 00:12:06,760 --> 00:12:08,520 the business models of their current borrowers. 265 00:12:09,240 --> 00:12:12,880 That's, I mean that's wild, but what if AI takes much longer to 266 00:12:12,880 --> 00:12:15,480 actually turn a profit? Well, then you face scenario 2. 267 00:12:16,160 --> 00:12:18,920 Enterprise AI monetization develops slowly. 268 00:12:19,240 --> 00:12:22,360 OK, If these newly financed AI infrastructure companies take 269 00:12:22,360 --> 00:12:25,400 years to become profitable, they simply won't have the operating 270 00:12:25,400 --> 00:12:27,920 cash flow to refinance their own maturing debt facilities. 271 00:12:28,200 --> 00:12:30,240 Right, when those initial three-year contracts expire. 272 00:12:30,240 --> 00:12:33,760 Exactly, and that triggers massive defaults on the new AI 273 00:12:33,760 --> 00:12:35,680 loans. So it's a structural trap. 274 00:12:36,080 --> 00:12:39,040 A total catch 22. If AI disrupts everything 275 00:12:39,040 --> 00:12:41,920 tomorrow, their massive legacy software portfolio burns down. 276 00:12:42,400 --> 00:12:45,120 If AI takes a decade to figure out how to be profitable, their 277 00:12:45,120 --> 00:12:47,640 new multi trillion dollar infrastructure portfolio burns 278 00:12:47,640 --> 00:12:49,720 down. That is the dual disruption 279 00:12:49,720 --> 00:12:52,360 paradox. The capital destruction risk 280 00:12:52,640 --> 00:12:54,400 exists on both ends of the timeline. 281 00:12:55,680 --> 00:12:58,320 Man, and because institutional investors are starting to 282 00:12:58,320 --> 00:13:01,680 realize they are caught in this trap, we are seeing the third 283 00:13:01,680 --> 00:13:05,600 flashing warning sign, right? Yes, a very real reaction in the 284 00:13:05,600 --> 00:13:08,680 market. Because in early 2026, something 285 00:13:08,680 --> 00:13:11,120 unprecedented happened with these private debt funds. 286 00:13:11,240 --> 00:13:13,760 Right for the first time on record, leading semi liquid 287 00:13:13,760 --> 00:13:16,320 private debt funds saw net capital outflows. 288 00:13:16,760 --> 00:13:19,840 Meaning they were returning more capital to withdrawing investors 289 00:13:19,840 --> 00:13:21,960 and they were raising through new subscriptions. 290 00:13:22,080 --> 00:13:23,920 Exactly. People wanted out. 291 00:13:24,360 --> 00:13:27,640 And just to unpack why that's so dangerous, these funds promise 292 00:13:27,640 --> 00:13:29,840 periodic liquidity to their investors, right? 293 00:13:30,280 --> 00:13:32,720 Meaning investors think they can pull their cash out when they 294 00:13:32,720 --> 00:13:36,160 want to, but the funds are holding fundamentally illiquid 295 00:13:36,160 --> 00:13:39,040 multi year compute loans. Which is a huge problem. 296 00:13:39,120 --> 00:13:41,520 Which is a huge problem, yeah. If investors keep asking for 297 00:13:41,520 --> 00:13:44,720 their money back, asset managers face a severe liquidity 298 00:13:44,720 --> 00:13:46,840 mismatch. They literally don't have the 299 00:13:46,840 --> 00:13:47,920 cash on hand. Right. 300 00:13:48,200 --> 00:13:51,520 They might be forced to halt new commitments or worse, liquidate 301 00:13:51,520 --> 00:13:55,160 the secondary loan portfolios at a steep loss just to generate 302 00:13:55,160 --> 00:13:57,200 the cash needed to pay off the withdrawing investors. 303 00:13:57,480 --> 00:14:00,360 Which means we are already seeing the secondary debt market 304 00:14:00,440 --> 00:14:02,400 start to aggressively price in these risks. 305 00:14:02,400 --> 00:14:04,680 Yes, it's happening right now. The report makes it clear this 306 00:14:04,680 --> 00:14:07,520 isn't just theoretical modeling anymore. 307 00:14:07,520 --> 00:14:10,440 The market is actively splitting into winners and losers based on 308 00:14:10,440 --> 00:14:12,520 real world constraints. Really is. 309 00:14:12,960 --> 00:14:15,880 Let's look at 2 massive multi billion dollar case studies from 310 00:14:15,880 --> 00:14:17,640 the report to see how this plays out in practice. 311 00:14:17,840 --> 00:14:19,840 OK, so the divergent here is sharp. 312 00:14:20,160 --> 00:14:23,200 Let's start with the debt financing for XA is compute 313 00:14:23,200 --> 00:14:24,040 clusters. Right. 314 00:14:24,040 --> 00:14:26,680 Elon Musk's company, right? In the secondary market, this 315 00:14:26,680 --> 00:14:29,760 specific debt is actually trading at a premium above face 316 00:14:29,760 --> 00:14:31,680 value. Meaning investors feel 317 00:14:31,680 --> 00:14:34,160 incredibly safe holding it to the point they'll actually pay 318 00:14:34,240 --> 00:14:35,480 extra for it. Exactly. 319 00:14:35,840 --> 00:14:37,680 And I'm guessing that has everything to do with corporate 320 00:14:37,680 --> 00:14:42,320 consolidation because SpaceX acquired XAI, meaning Spacex's 321 00:14:42,320 --> 00:14:45,120 broader aerospace balance sheet is essentially Co signing the 322 00:14:45,120 --> 00:14:46,360 loan. That's it exactly. 323 00:14:46,560 --> 00:14:49,760 It acts as a backstop. Spacex's robust corporate 324 00:14:49,760 --> 00:14:53,280 balance sheet absorbs the tenant risk OK, effectively converting 325 00:14:53,280 --> 00:14:57,080 what would normally be a highly speculative project debt into a 326 00:14:57,240 --> 00:14:59,040 high performing stable credit instrument. 327 00:14:59,320 --> 00:15:01,840 So having a giant stable corporate parent makes the debt 328 00:15:01,840 --> 00:15:03,720 safe? Yes, but contrast that with a 329 00:15:03,720 --> 00:15:06,000 second case study, Project Jupiter. 330 00:15:06,240 --> 00:15:09,000 Yes, this was a joint venture between Oracle and Open AI, 331 00:15:09,360 --> 00:15:11,760 right? An $18 billion debt facility to 332 00:15:11,760 --> 00:15:16,480 build a massive 2.45 GW data center campus in New Mexico. 333 00:15:16,680 --> 00:15:19,560 A massive undertaking. And this was backed by what 334 00:15:19,560 --> 00:15:22,680 lawyers call hell or high water lease covenants from Oracle. 335 00:15:22,880 --> 00:15:25,120 Which is supposed to be the absolute strongest legal 336 00:15:25,120 --> 00:15:27,600 guarantee available, right? It means Oracle was 337 00:15:27,600 --> 00:15:30,000 contractually obligated to service the lease payments 338 00:15:30,040 --> 00:15:33,320 regardless of operational readiness, equipment failure, 339 00:15:33,320 --> 00:15:35,400 basically anything. Come hell or high water, you are 340 00:15:35,400 --> 00:15:36,800 paying this loan. Exactly. 341 00:15:37,120 --> 00:15:40,120 Yet despite an ironclad contract from one of the most established 342 00:15:40,120 --> 00:15:43,280 tech companies on earth, the report says this $18 billion 343 00:15:43,280 --> 00:15:45,240 debt is heavily distressed. Heavily. 344 00:15:45,440 --> 00:15:48,920 It's trading at like 89 to 91 cents on the dollar, and 345 00:15:48,920 --> 00:15:51,640 Oracle's credit default swap spreads surged. 346 00:15:51,640 --> 00:15:54,040 They absolutely spiked. Let's just translate that last 347 00:15:54,040 --> 00:15:56,680 point for you listening. That basically means the cost of 348 00:15:56,680 --> 00:15:59,080 buying financial insurance against Oracle failing to pay 349 00:15:59,080 --> 00:16:01,560 its debts just skyrocketed in the markets, right? 350 00:16:01,640 --> 00:16:04,760 Yes, the market suddenly saw a real risk of Oracle defaulting 351 00:16:04,760 --> 00:16:08,040 on this specific project. But wait, if the contract 352 00:16:08,040 --> 00:16:11,160 legally bound them to pay no matter what, why did the debt 353 00:16:11,160 --> 00:16:13,480 tank? Because the digital revolution 354 00:16:13,480 --> 00:16:15,680 hit a very tangible physical wall. 355 00:16:15,760 --> 00:16:18,280 What happened? Environmental regulations and 356 00:16:18,280 --> 00:16:21,160 local opposition successfully blocked the construction of a 357 00:16:21,160 --> 00:16:23,440 critical 17 mile natural gas pipeline. 358 00:16:23,840 --> 00:16:27,240 A pipeline. Yeah, and that pipeline was 359 00:16:27,240 --> 00:16:29,840 strictly required to power the site's generators. 360 00:16:30,640 --> 00:16:33,360 Oh man. Because they physically couldn't 361 00:16:33,360 --> 00:16:36,600 get power to the facility, Oracle delivered a formal force 362 00:16:36,600 --> 00:16:38,160 majeure notice to the developers. 363 00:16:38,400 --> 00:16:42,720 Force majeure, an unforeseeable, unavoidable event that legally 364 00:16:42,720 --> 00:16:46,560 frees a party from liability. Precisely by invoking it, Oracle 365 00:16:46,560 --> 00:16:49,320 legally delay their rent and carry cost obligations. 366 00:16:49,440 --> 00:16:51,680 Wow. And the secondary debt market 367 00:16:51,680 --> 00:16:54,400 just panicked immediately. Of course, without that 368 00:16:54,400 --> 00:16:58,280 pipeline, there was massive uncertainty over when or even if 369 00:16:58,600 --> 00:17:01,840 that $18 billion asset would ever generate a single dollar of 370 00:17:01,840 --> 00:17:03,960 cash flow. I just have to pause on the 371 00:17:03,960 --> 00:17:05,960 sheer irony here it. Is pretty rich. 372 00:17:06,280 --> 00:17:10,520 You have a trillion dollar digital AI revolution, right? 373 00:17:10,520 --> 00:17:13,760 The pinnacle of complex human technology, and it gets 374 00:17:13,760 --> 00:17:17,720 completely derailed by 17 miles of physical pipe in the dirt. 375 00:17:18,079 --> 00:17:19,599 It is the ultimate reality check. 376 00:17:20,040 --> 00:17:22,680 It absolutely doesn't matter how bulletproof your financial 377 00:17:22,680 --> 00:17:26,920 contract is or how sophisticated your generative AI model is if 378 00:17:26,920 --> 00:17:29,440 you physically cannot plug the server into the wall. 379 00:17:29,760 --> 00:17:32,240 Investment grade tenant commitments simply cannot 380 00:17:32,240 --> 00:17:36,080 neutralize execution risks when basic physical utility access 381 00:17:36,080 --> 00:17:38,280 fails. Now, if you were listening to 382 00:17:38,280 --> 00:17:41,160 this and keeping track, we're talking about billions in 383 00:17:41,160 --> 00:17:44,600 distressed, mispriced debt. We've got forced redemptions 384 00:17:44,600 --> 00:17:47,600 from panicked investors. We've got structural market 385 00:17:47,600 --> 00:17:50,640 paradoxes where lenders lose whether AI succeeds or fails. 386 00:17:50,640 --> 00:17:52,640 It's a lot. And we have massive projects 387 00:17:52,640 --> 00:17:55,560 failing over simple utility blockages. 388 00:17:56,360 --> 00:17:59,360 It leads to the terrifying question, is this 2008 all over 389 00:17:59,360 --> 00:18:00,760 again? Naturally, people are asking. 390 00:18:00,760 --> 00:18:02,520 That right. Are we looking at a systemic 391 00:18:02,520 --> 00:18:05,560 banking crisis just swapping out subprime mortgages for subprime 392 00:18:05,560 --> 00:18:07,160 GPUs? Well, the social capital 393 00:18:07,160 --> 00:18:10,120 analysis directly addresses this and their conclusion is 394 00:18:10,120 --> 00:18:12,320 definitive, no? OK. 395 00:18:12,800 --> 00:18:15,800 Yeah, this is an isolated credit bubble, not a systemic banking 396 00:18:15,800 --> 00:18:17,360 contagion. Why? 397 00:18:17,840 --> 00:18:20,680 Because today's market has significant structural cushions 398 00:18:20,680 --> 00:18:22,960 that simply did not exist in the lead up to 2008. 399 00:18:23,200 --> 00:18:24,400 OK. What kind of cushions are we 400 00:18:24,400 --> 00:18:26,760 talking about? Why won't this take down the 401 00:18:26,760 --> 00:18:29,400 broader economy? Well, first you have third party 402 00:18:29,400 --> 00:18:31,920 industrial guarantees. These fundamentally alter 403 00:18:31,920 --> 00:18:34,160 recovery expectations. How so? 404 00:18:34,160 --> 00:18:37,880 Give me an example. Take that $35 billion Anthropic 405 00:18:37,880 --> 00:18:41,560 deal we discussed earlier, right Broadcom, the company that Co 406 00:18:41,560 --> 00:18:44,760 designed the TPU chips, they agreed to provide residual value 407 00:18:44,760 --> 00:18:46,880 guarantees on the senior debt tranches. 408 00:18:47,320 --> 00:18:49,960 Meaning the investors who are first in line to get paid back 409 00:18:50,160 --> 00:18:52,160 the senior tranches, they actually have a safety net. 410 00:18:52,280 --> 00:18:55,040 Exactly. If the AI model fails and the 411 00:18:55,040 --> 00:18:58,880 SPV goes bankrupt, Broadcom backstops the residual value of 412 00:18:58,880 --> 00:18:59,840 the physical silicon. Oh. 413 00:19:00,440 --> 00:19:01,680 That's smart. Right. 414 00:19:01,760 --> 00:19:03,960 It creates A defined floor for liquidity. 415 00:19:04,080 --> 00:19:07,480 OK. But the second and arguably much 416 00:19:07,480 --> 00:19:10,680 more important factor is the operational leverage profile. 417 00:19:10,880 --> 00:19:14,120 The leverage, right? Prior to the 2008 subprime 418 00:19:14,120 --> 00:19:17,160 crisis, structured investment vehicles operated with leverage 419 00:19:17,160 --> 00:19:21,040 ratios that frequently exceeded 30 to one against those subprime 420 00:19:21,040 --> 00:19:24,840 mortgages. 30 to 1, yes. And that was funded heavily by 421 00:19:24,840 --> 00:19:28,560 traditional commercial banks. So for everyone actual dollar 422 00:19:28,560 --> 00:19:31,480 they had, they were borrowing and betting $30. 423 00:19:31,600 --> 00:19:33,320 Exactly. That's terrifying. 424 00:19:33,600 --> 00:19:36,720 It was, but today's private credit funds are engineered 425 00:19:36,720 --> 00:19:39,560 entirely differently. They maintain fund level 426 00:19:39,560 --> 00:19:42,600 leverage ratios between just one to 1:00 and 2:00 to 1:00. 427 00:19:43,080 --> 00:19:45,600 Oh wow, so way. Lower, drastically lower. 428 00:19:46,080 --> 00:19:48,920 And crucially, this is financed primarily through long term 429 00:19:48,920 --> 00:19:50,480 limited partner equity commitments. 430 00:19:50,800 --> 00:19:53,360 Which means who exactly is putting up the money? 431 00:19:53,640 --> 00:19:56,520 We're talking pension funds, sovereign wealth funds and 432 00:19:56,520 --> 00:19:58,240 massive institutional allocators. 433 00:19:58,880 --> 00:20:01,320 OK, so it's the difference between a desperate retail 434 00:20:01,320 --> 00:20:03,880 trader taking out a massive second mortgage to bet on a 435 00:20:03,880 --> 00:20:07,400 stock and a syndicate of multibillion dollar pension 436 00:20:07,400 --> 00:20:10,120 funds betting their spare cash. That's a perfect analogy. 437 00:20:10,600 --> 00:20:13,720 If the AI bet fails, those pension funds take a massive 438 00:20:13,720 --> 00:20:15,680 loss, which certainly hurts their returns. 439 00:20:15,720 --> 00:20:18,640 Right, it's bad news for them. Bad news for them, but the 440 00:20:18,640 --> 00:20:20,400 global banking system doesn't collapse. 441 00:20:20,840 --> 00:20:23,000 Got it. Credit distress and massive 442 00:20:23,000 --> 00:20:25,680 project write downs in this AI infrastructure space will 443 00:20:25,680 --> 00:20:29,160 decimate the returns for those specific limited partners, but 444 00:20:29,160 --> 00:20:31,960 it will not 'cause systemic commercial bank insolvencies. 445 00:20:32,040 --> 00:20:34,680 The damage is largely contained within the private credit 446 00:20:34,960 --> 00:20:36,000 ecosystem. Precisely. 447 00:20:36,160 --> 00:20:39,120 So Synthesizing everything we've looked at in this deep dive, 448 00:20:39,600 --> 00:20:42,760 we've tracked this historical shift from equity funding to 449 00:20:42,760 --> 00:20:45,040 trillions of dollars in structured shadow debt. 450 00:20:45,200 --> 00:20:48,320 Yep, we've uncovered the incredibly dangerous duration 451 00:20:48,320 --> 00:20:51,520 mismatch of funding rapidly depreciating chips on short term 452 00:20:51,520 --> 00:20:55,160 contracts using long term loans. The smartphone mortgage. 453 00:20:55,200 --> 00:20:58,080 Exactly. We mapped out that damned if you 454 00:20:58,080 --> 00:21:02,160 do dual disruption paradox where legacy SAS portfolios might burn 455 00:21:02,160 --> 00:21:06,080 down to build the AI future. And finally, we've seen why, 456 00:21:06,080 --> 00:21:09,160 despite all these flashing red lights, this isolated bubble 457 00:21:09,160 --> 00:21:10,720 isn't going to crash your local bank. 458 00:21:11,120 --> 00:21:14,400 It's contained, but just because it won't crash the banking 459 00:21:14,400 --> 00:21:17,000 system doesn't mean A reckoning isn't coming for the tech 460 00:21:17,000 --> 00:21:17,960 sector. Right. 461 00:21:17,960 --> 00:21:21,720 When does the bill come due? The report makes it clear that 462 00:21:21,720 --> 00:21:24,880 the true macroeconomic stress test is looming just over the 463 00:21:24,880 --> 00:21:28,760 horizon in 2027 and 2028. Which is barely a couple of 464 00:21:28,760 --> 00:21:30,040 years away. Exactly. 465 00:21:30,440 --> 00:21:35,040 That is when the $1.3 trillion annual hyperscaler CapEx bill 466 00:21:35,040 --> 00:21:37,600 comes due. That is when entirely new, 467 00:21:37,840 --> 00:21:40,600 vastly superior chip architectures will arrive, 468 00:21:40,920 --> 00:21:43,520 instantly depreciating the collateral backing all these 469 00:21:43,520 --> 00:21:44,800 current loans. Oh wow. 470 00:21:44,880 --> 00:21:47,600 And most importantly, that is when those initial three-year 471 00:21:47,600 --> 00:21:51,040 compute lease agreements expire. The end of the road. 472 00:21:51,040 --> 00:21:54,440 Lenders are going to be forced to mark their collateral to true 473 00:21:54,440 --> 00:21:57,400 stark reality. The public narrative exuberance 474 00:21:57,400 --> 00:22:00,000 is finally going to collide with hard mathematical reality. 475 00:22:00,000 --> 00:22:02,160 Exactly right. And as we wrap up today, I 476 00:22:02,160 --> 00:22:04,520 really want to leave you listening with one final thought 477 00:22:04,520 --> 00:22:07,200 to Mull over connecting back to the most surprising piece of our 478 00:22:07,200 --> 00:22:08,960 discussion, honestly, the pipeline. 479 00:22:09,160 --> 00:22:12,320 The pipeline. If tangible physical constraints 480 00:22:12,320 --> 00:22:16,200 like a single 17 mile natural gas pipeline can completely 481 00:22:16,200 --> 00:22:19,760 derail an $18 billion project backed by some of the most 482 00:22:19,800 --> 00:22:23,840 powerful companies on Earth, what happens when the almost 483 00:22:23,840 --> 00:22:28,480 unfathomable energy demands of the 2027 AI build out start 484 00:22:28,480 --> 00:22:31,160 competing directly with the power grids of everyday cities? 485 00:22:31,520 --> 00:22:34,360 The physical infrastructure is the ultimate undeniable 486 00:22:34,360 --> 00:22:36,680 bottleneck. Will the ultimate cap on a is 487 00:22:36,680 --> 00:22:39,880 explosive growth be financial, or will it simply be electrical? 488 00:22:40,200 --> 00:22:42,280 It's definitely something to think about every time you plug 489 00:22:42,280 --> 00:22:43,760 in your devices. Absolutely. 490 00:22:44,040 --> 00:22:46,440 Thank you so much for joining us as we impact the hidden 491 00:22:46,440 --> 00:22:49,520 mechanics of the AI boom. It's been a fascinating shift to 492 00:22:49,520 --> 00:22:50,760 analyze. Thanks for having me.