शैडोइंग अभ्यास: The AI Bubble Survives on $1.65 Trillion in Hidden Debt - YouTube के साथ अंग्रेजी बोलना सीखें
C1
⏸ रुका हुआ
You've seen the AI headlines.
201 वाक्य
यदि वाक्य बहुत छोटे या लंबे हों, तो समायोजित करने के लिए Edit पर क्लिक करें।
1
You've seen the AI headlines.
2
In fact, you've probably seen so many of them that the numbers have stopped meaning anything.
3
Every week it's another hundred billion, another mega campus, another record.
4
The numbers have gotten so big, they've stopped registering at all.
5
They're just something you scroll past on Instagram, think, oh wow, and three seconds later, you're back to watching a guy make pasta.
6
But we shouldn't be moving on so fast, because this year alone, Big Tech's on track to spend more than $700 billion building out AI.
7
That's roughly 2.5 cents of every dollar the entire United States economy produces.
8
The CEO of Nvidia, Jensen Huang, is calling AI the largest infrastructure build-out in human history.
9
But nobody's asking the question sitting right in front of our eyes.
10
How are we actually paying for this sh?
11
The answer to that starts with one number.
12
$1.65 trillion.
13
That's what one study found just five tech companies owe in obligations that never appear on their balance sheets.
14
Meanwhile the same five companies report about $1.35 trillion of debt on the books.
15
So the hidden pile of debt is not only bigger than the visible one, it's much bigger.
16
Take Meta for example.
17
Their off-balance sheet debt is nearly triple the reported debt.
18
Moving that much money out of sight isn't a trick you can invent overnight.
19
It takes a system, one with a name, and it's a name Wall Street has spent the last two decades trying very hard to forget.
20
So we have to go back to the summer of 2007, right before the global financial crisis.
21
An economist at PIMCO, one of the largest asset managers on earth, stands up at the Federal Reserve's annual symposium in Jackson Hole and gives a warning.
22
His name is Paul McCulley, and he's growing increasingly scared of a second banking system that's growing in the dark.
23
It's institutions that borrow like banks and lend like banks, but aren't banks.
24
So no deposit insurance, no Fed safety net, and none of the regulation written for banks.
25
McCulley gave it a name, the shadow banking system, which he describes as the whole alphabet soup of levered up non-bank investment conduits, vehicles, and structures.
26
Now if non-bank conduits, vehicles, and structures sounds like word salad, here's the plain English version of how the shadow banking system works.
27
Let's say we have a company called the Acme Corporation.
28
They want to borrow a mountain of money, but they don't want the debt showing up on their own books.
29
The solution is Acme creates a second company called a special purpose vehicle, or special purpose entity.
30
On paper, Acme and the SPV are completely separate.
31
They have their own names, own bank accounts, and most importantly, their own debts.
32
The SPV takes on a bunch of debt and uses that borrowed money to go out and buy the assets Acme needs.
33
Then the SPV leases the assets back to Acme for them to use.
34
The SPV borrows the money and owns the asset.
35
Then the original company gets to use it without having the debt on their books.
36
It's like opening a credit card in your cousin's name.
37
You're the one swiping it.
38
You're the one enjoying everything it buys.
39
But when the bank asks about your debts, your answer is technically zero.
40
In 2007, the shadow banking system was built on piles of mortgages.
41
Today, it's data centers.
42
The AI build-out runs on financing machinery from the same family that broke the world in 2008.
43
But the shadows aren't where this story starts.
44
They're where the money comes from when everywhere else runs out.
45
And before we get to that, a quick pause.
46
Because time is your most valuable asset.
47
If you spend just 3 hours a day typing, going from 40 words per minute to 70 saves you roughly 50 minutes daily, more than 300 hours per year.
48
But what if you could type at 220 words per minute?
49
How much time could you get back then?
50
Which brings us to the sponsor of today's video, Whisperflow.
51
Whisperflow is a voice-to-text tool that actually works.
52
Take a look.
53
Hi Ananya.
54
Following up on the question from Thursday, I moved part of the position on the 14th, sorry, uh, I mean the 15th, and I want to confirm whether that triggers the wash sale rule,
55
given I bought back in within 20 days.
56
Sorry, I mean 30 days.
57
Whisperflow cleaned up every mistake, removed the filler words, and formatted the whole thing, all in about 15 seconds.
58
But with Whisperflow, it's not just about speed, it's also the accuracy.
59
Here's the same words spoken using two different tools.
60
Whisperflow on the left, built-in voice on the right.
61
Whisperflow stripped out my mistakes, handled the corrections, and got the spelling right, even the tickers.
62
Whisperflow turns your speech into clear, polished writing across any app, any device.
63
If you're interested, Whisperflow is free to use, and you can get one month of Whisperflow Pro with unlimited words for free using the link below, promo code casual.
64
Thank you to Whisperflow for sponsoring this video.
65
Now back to the money behind the AI build-out and the family of financing it comes from.
66
The same family that broke the world in 2008.
67
Which matters because the thing that needs the money right now is the largest infrastructure build-out in human history.
68
No single source of money on Earth is deep enough to fund it alone, so big tech started knocking on every door that might have money behind it.
69
But it started from the most obvious source, their own vaults.
70
On the surface, these companies are the safest borrowers on Earth.
71
According to PIMCO, the highest-rated names in this group carry net debt of just 0.04 times their earnings, the lowest of any sector in the United States.
72
So the plan started off simple, let the cash machine pay for the build-out.
73
But then, the machine started getting eaten.
74
For most of the last decade, these companies spent roughly 40 to 50 cents of every dollar of cash their businesses produced on capital expenditures,
75
the formal name for reinvesting cash back into your business.
76
This year, PIMCO estimates, that number hits 94 cents.
77
That's tech companies running their entire cash engine to finance the AI build-out.
78
So it's not the vault is funding the build-out anymore.
79
It's the build-out is draining the vault.
80
And when these companies ran out of cash to reinvest, they hit the bond markets, at a pace we've never seen before.
81
From 2020 to 2024, the five biggest hyperscalers issued on average $28 billion a year of bonds.
82
In 2025, the same five biggest hyperscalers issued more than $100 billion, more than four times their average pace over the previous five years.
83
When 2026 happened, in just the first five months of this year, they've already issued another $159 billion.
84
Meaning, in five months, they borrowed more than they had in the entire previous five years combined.
85
The technology sector now accounts for nearly 18% of the total U.S investment-grade debt outstanding.
86
It's gotten so out of control that the International Monetary Fund is now flagging it.
87
Tobias Adrian, a senior official at the IMF and former senior vice president at the Federal Reserve Bank of New York,
88
laid it out simply, what is quite worrisome from a financial stability perspective is
89
that the major tech firms are starting to leverage up themselves.
90
And when the US market got saturated with tech's debt, they went international.
91
In February, Alphabet raised about $32 billion across four currencies in under 24 hours,
92
including a 100-year bond that was roughly 10 times oversubscribed.
93
Overnight, these These companies became top bond issuers in currencies where they had zero bonds a year earlier.
94
A bond manager at Rathbones explained why, Big tech had to issue in all currencies, as the US dollar market, which is the largest credit market globally, can't absorb it.
95
Which means the deepest capital market on earth cannot absorb what these tech companies need, which forced them to look somewhere else.
96
And somewhere else is where this story stops being normal, because you can only issue so much debt on your books.
97
Which brings us back to the special purpose vehicle and the shadow banking system from earlier.
98
The main difference between 2008 and today is that the shadow banking system today is hiding in plain sight.
99
Everyone's missing it because of where it lives.
100
It's buried in the footnotes of public filings, and the only people who read footnotes are finance geeks.
101
Well, and me.
102
If you open the footnotes, the first thing you notice is that it's crowded down there.
103
I could read out names and numbers for an hour, and you'd be no closer to understanding how any of it works.
104
So instead, we'll zoom in on one deal, the biggest ever done, the blueprint every deal after it copied, the largest project financing ever completed as an investment-grade bond,
105
$27 billion, all riding on a special purpose vehicle named after a New Orleans fried pastry.
106
The company is called Beignet Investor.
107
It's the exact shadow banking machine from earlier running live.
108
It exists for exactly one purpose, to own a data center campus called Hyperion, a multi-gigawatt site in Richland Parish, rural Louisiana.
109
And Hyperion is a Meta project.
110
Meta announced it.
111
Meta designs it.
112
Meta builds it.
113
Meta operates it.
114
Meta is the only tenant it will ever have.
115
But the $27 billion of debt that built the Hyperion Data Center campus belongs to the pastry.
116
Here's how it works.
117
Blue Owl Capital, a private investment firm, creates the special purpose vehicle called Beignet Investor and puts up 80% of the equity.
118
Meta takes the other 20% and signs on as the the building's only tenant, then the pastry goes borrowing.
119
Beignet issues $27.3 billion of investment-grade bonds.
120
The bondholders, who are some of the biggest money managers on earth, like BlackRock and Pimco, wire in the cash.
121
Beignet takes that cash and builds Hyperion.
122
Then the final step, Meta moves in and pays rent.
123
And that rent is what pays the bondholders back year after year all the way to 2049.
124
So it's Meta's data center, Meta's machines, Meta's engineers incite it.
125
But Meta legally owns just 20% of it, while Blue Owl owns the other 80%.
126
Why?
127
It's because of a single phrase in the accounting rules.
128
At 20%, Meta is not the primary beneficiary of the venture.
129
And if you're not the primary beneficiary, the venture's $27 billion of debt is not your debt.
130
So it vanishes from your balance sheet, legally.
131
But there's also more hiding in the fine print of Meta's own filings.
132
On page 106 of Meta's annual report, one sentence hidden at the bottom reads,
133
We have provided residual value guarantees with an aggregate threshold of approximately $28 billion that decreases over time.
134
If that means nothing to you, don't worry.
135
I had to read it four times before it landed.
136
It basically just means Meta's guaranteeing the building.
137
If Meta ever walks away and Hyperion is worth less than a guaranteed floor, Meta Meta writes the check for the difference.
138
Say the campus drops to $20 billion against the $28 billion floor, Meta now has to write an $8 billion check to make the bondholders whole.
139
So to Wall Street, Meta gets to say, we don't own it, we didn't borrow the $27 billion.
140
This isn't our liability.
141
To the bondholders, Meta gets to say, don't worry, if this thing loses value when we leave, we've got you covered for up to $28 billion.
142
It's the same thing, but Meta's able to sell two different stories.
143
Then there's the cherry on top.
144
The bonds issued by Beignet to finance the build-out of Hyperion run to 2049.
145
But Meta's lease doesn't even start until 2029, and it renews in four-year chunks.
146
So it's a 24-year loan, resting on a four-year lease.
147
And remember what this building is?
148
It's a hyper-customized supercomputing site built around Meta's own hardware in rural Louisiana.
149
So if Meta ever walks, the bondholders are left holding a very large, very specific building for the next two decades.
150
But Beignet didn't stay as a one-off deal.
151
It became a template.
152
Let me introduce you to Project Sopapia.
153
And no, that's not a joke.
154
Last week, BlackRock started marketing the sequel to Project Beignet, Project Sopapia.
155
More than $12 billion of bonds for Meta's newest 1-gigawatt AI data center in El Paso, Texas, with JP Morgan and Morgan Stanley running the sale.
156
Same structure, same 80-20 split, but different pastry.
157
On their own, two deals is just two deals.
158
It doesn't prove much much.
159
But what's scary is the nine-month trend that followed.
160
Estimates show the whole off-book pile of debt, across the five companies, has grown roughly eight times over since 2022.
161
The law firm Quinn Emanuel counted more than $120 billion moved off balance sheets in under two years.
162
Moody's found Big Tech signed legal contracts pledging nearly $1 trillion for future AI data center leases.
163
But over $660 billion of those obligations is completely invisible on their main balance sheets today.
164
Morgan Stanley's estimates put the total off-balance sheet exposure across the industry at roughly $1.8 trillion.
165
So it's not just meta, we're seeing it across the tech industry.
166
And here's what makes this a deeper, systemic problem.
167
It's the same small club of lenders on every deal.
168
When five borrowers use one structure funded by one handful of lenders, the risk stops being diversifiable.
169
It's the same bet, written five times.
170
The last two times a financing structure standardized this fast across
171
an entire industry were telecom vendor financing from 1998 to 2000 and mortgage securitization from 2004 to 2007.
172
A lawyer at Quinn Emanuel, a man who spent years on the litigation that followed 2008, described watching this unfold with two words, déjà vu.
173
So there's the answer to the question we opened with, how is the largest infrastructure build-out in human history being paid for?
174
First with cash, until the machines started eating 94 cents of every dollar.
175
Then with bonds, until the deepest capital market on earth couldn't absorb them.
176
Then now in the footnotes, where the hidden pile of debt has quietly outgrown the public facing balance sheet totals.
177
So where does that leave us?
178
Honestly, I've spent weeks inside these filings, and I have no idea how this ends.
179
The story gets crazier every week.
180
But I can tell you what has to go right.
181
Because this isn't really an investment anymore.
182
It's the largest bet in history.
183
And the bet has four legs.
184
First, the revenue has to multiply, not grow, multiply.
185
AI revenue is estimated at around $110 billion a year, against more than $700 billion of spending.
186
Second, the returns have to stop shrinking.
187
By one industry analysis, the return on every new dollar the cloud giants invest has already slid from around 40% toward 20,
188
with projections pointing toward 10 if the spending keeps up.
189
Third, the borrowing window has to stay open.
190
Morgan Stanley tallies the build out at roughly $2.9 trillion through 2028.
191
The company's own cash flows cover about $1.4 trillion of it, which leaves a $1.5 trillion hole between the plan and the money.
192
And fourth, the grid has to absorb all of it.
193
And the largest power market in America is already pinned at its price cap.
194
But that's a story for a different day, and if you want me to go into this horror story, drop a comment, and I will.
195
None of this AI stuff is a secret.
196
It's just hiding in public filings, auction records, and analyst reports the stuff nobody reads except the nerds in Patagonia Vest.
197
And that's what this channel is.
198
I do all the reading, and you get all the parts that matter.
199
If that's worth something to you, hit subscribe.
200
Because when you step back and look at everything that has to go right for AI, this stops looking like an investment into an emerging technology, and it starts looking like a parlay.
201
Four legs stacked on top of each other, and every single leg has to hit for this to pay off.
ऐप डाउनलोड करें
Everything you need to speak fluently
AI PronunciationScore every sentence
IPA PracticeMaster every sound
VocabularyBuild your word bank
Vocab GameLearn while playing
इस पाठ के बारे में
आप "The AI Bubble Survives on $1.65 Trillion in Hidden Debt" के साथ Shadowing तकनीक का उपयोग करके अपनी अंग्रेजी का अभ्यास कर रहे हैं।
शैडोइंग तकनीक क्या है?
शैडोइंग (Shadowing) एक विज्ञान-समर्थित भाषा सीखने की तकनीक है जो मूल रूप से पेशेवर दुभाषिया प्रशिक्षण के लिए विकसित की गई थी। विधि सरल लेकिन शक्तिशाली है: आप मूल अंग्रेज़ी ऑडियो सुनते हैं और तुरंत इसे ज़ोर से दोहराते हैं — जैसे वक्ता की छाया 1-2 सेकंड की देरी से। शोध से पता चलता है कि यह उच्चारण सटीकता, स्वर, लय, जुड़ी हुई ध्वनियाँ, सुनने की समझ और बोलने की प्रवाहशीलता में काफ़ी सुधार करता है।