YT YouTube2Text

YouTube2Text → Transcripts → Why Wall Street is Ignoring Big Tech's Debt

Why Wall Street is Ignoring Big Tech's Debt — Transcript

by Patrick Boyle · 5,015 words · 748 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00A few weeks ago, Nick Aasia reported
0:02that the five biggest US tech companies
0:05are carrying $1.65 trillion of debt that
0:09doesn't appear anywhere on their balance
0:10sheets. Not the debt that you can see, a
0:14second larger pile hidden behind it. A
0:17few days later, the Financial Times
0:18found another 50 billion in leases that
0:22Nvidia had signed for a single data
0:24center in Texas, stuffed full of its own
0:27chips. a commitment nobody had known
0:30about. And the number keeps moving.
0:33Nicki did their count before most of
0:35these companies had even reported
0:37earnings. And when they did report days
0:39ago, three of them alone signed nearly
0:42$900 billion of new AI commitments in a
0:46single quarter. So 1.65 trillion is
0:50already an underestimate. Which is why
0:53if you spend any time on financial
0:55YouTube, you already know what people
0:58are calling this. The word being thrown
1:00around is Enron. Commentators, the ones
1:03with big social media followings and no
1:05obvious background in accounting, have
1:08looked at these numbers and concluded
1:10it's Enron all over again. So, I've been
1:14practicing my shocked face in the
1:15mirror. It turns out to be surprisingly
1:18hard to hold that frozen open-mounted
1:21pointing at a redline expression while
1:24also looking like you understand what a
1:26lease is. Because here's the question
1:28this video is actually about. Is any of
1:31that true? Is this fraud? The real
1:34numbers being hidden from investors the
1:36way Enron hit them right up until the
1:39whole thing fell apart. Or is it
1:41something much more boring and much more
1:44interesting? Let's find out. First
1:48though, because a lot of you weren't
1:50following the financial news in 2001, a
1:53quick word on Enron since the entire
1:56accusation rests on it. Enron was an
1:59American energy giant that turned out to
2:02be a fraud. It had been hiding enormous
2:05debts and losses in a web of secret
2:08offthe-books entities. The accounts
2:10investors could see were essentially a
2:13fiction. When it unraveled, the company
2:16collapsed in a matter of weeks. It took
2:18down Arthur Anderson, one of the five
2:21biggest accounting firms in the world
2:23and wiped out the retirement savings of
2:26thousands of its own employees. It is
2:28still the definitive corporate
2:30accounting fraud. So, when someone
2:32points at big tech and says Enron, they
2:35aren't complaining about confusing
2:37bookkeeping. They're alleging deliberate
2:40fraud on a criminal scale. That's the
2:43charge that we're going to test. When
2:45you see a headline claiming that tech
2:48giants are hiding over a trillion
2:50dollars in debt, it's natural to assume
2:52a crime is being committed. But when you
2:56dig a bit deeper, a lot of this debt
2:58turns out to be long-term purchase
3:00agreements for graphics cards and leases
3:02on data centers that haven't been built
3:05yet. Under standard accounting rules, if
3:08the goods haven't been delivered or if
3:10the building isn't running, you don't
3:12record it as a liability on the balance
3:14sheet. You disclose it in the footnotes.
3:18When you sign a 2-year phone contract,
3:20you've committed to paying the network
3:23something like $50 a month for the next
3:2524 months. That's a real obligation. You
3:29can't just stop. And if you added it up,
3:32you're on the hook for over $1,000.
3:35But you don't sit down and record a
3:37$1,200 liability on your personal
3:40balance sheet the day you sign. You pay
3:42for it month by month as you use it. The
3:45tech companies are doing the exact same
3:47thing, just with more zeros. Instead of
3:50a phone contract, it's a 15-year lease
3:52on a data center in Ohio. For decades,
3:56financial commentators have been
3:58complaining about tech companies
3:59hoarding cash that they use their cash
4:02flow to buy back shares instead of
4:04investing in anything new. Now, these
4:06same companies are issuing securities to
4:09invest in new infrastructure. And the
4:11same commentators have found a way to be
4:14unhappy about that, too. Which raises
4:16the question, what does it actually
4:19signal when a company chooses to borrow
4:21instead of raising equity? because it
4:24tells you quite a lot. But before I dig
4:26into that, let me tell you about this
4:28week's video sponsor, Printify. Printify
4:31is a print ondemand platform built for
4:34starting a side hustle from scratch with
4:36no inventory to buy and no cash upfront.
4:40You come up with a design, they handle
4:42the printing and shipping, and you never
4:44need to deal with the inventory. There
4:47are no upfront costs, no warehouse, no
4:50boxes of unsold mugs stacked up in your
4:52garage. The catalog runs to over 1,300
4:57products. And because printing happens
4:59through a network of local facilities
5:01across the US and Europe, your order
5:04gets made near wherever it's shipping,
5:07which keeps the fees down and the
5:09delivery times reasonable. It connects
5:12directly to your online sales platform.
5:15So, you're just adding a product line to
5:17a store you might already run rather
5:20than building one from nothing. If
5:22design isn't your thing, they've got AI
5:25tools built in to help you with design.
5:28Use my promo code Patrick to get 20% off
5:31your first order. The offer is valid for
5:34the first 100 customers only, so be
5:37quick. Click the link in the description
5:39to create your Printify account for free
5:42and start your own business today.
5:45Investors pay close attention to how a
5:48company pays for things. This is because
5:50the choices they make send clear signals
5:53to the market. Here's the intuition. If
5:56you actually thought you'd build a
5:58machine that turns $1 into five, you
6:00wouldn't sell half of it to strangers to
6:02raise the money to build it. You'd
6:04instead try to borrow the money, build
6:07the machine, keep the entire $5 profit,
6:10and pay off the loan. You want to keep
6:12all of the ownership stake yourself in
6:15such a high conviction investment. You
6:18only want to sell a percentage of the
6:20business when you're less sure about the
6:22likelihood of it making a lot of money.
6:24So, debt isn't always a bad sign.
6:27Borrowing to build tends to signal that
6:30management thinks the return on
6:32investment is worth keeping for the
6:34existing owners. Although, and this is
6:37where it gets interesting, these firms
6:39are also raising equity. In June,
6:42Alphabet completed the largest equity
6:45raise in corporate history, almost $85
6:48billion, anchored by a $10 billion check
6:51from Berkshire Hathaway. Berkshire might
6:54be the last name you'd expect on that
6:57list. A firm that generally regards
7:00buying back its own shares as more
7:02sensible than funding somebody else's
7:04moonshot. And a firm which drives a
7:07notoriously hard bargain. It reportedly
7:10bought its Alphabet stock at around a 6%
7:13discount to the market price because of
7:16course it did. This is not a firm that
7:18overpays for a story. But they decided
7:22that the AI buildout was worth a $10
7:25billion investment. Anyway, when you see
7:27big tech raising money by every route
7:30available all at once, record debt,
7:33record equity, convertibles, the lot,
7:36the signal isn't in which one they
7:38picked. It's in the scale of the capital
7:41raise itself. You don't raise capital
7:44like a company fighting for its life
7:46unless you think there's something on
7:48the other side worth the fight. Whether
7:50they're right about that is the rest of
7:53this video, but it's hard to argue
7:55they're hiding the spending when they
7:58announced part of it in the largest
8:00stock offering ever filed. And a lot of
8:03this resolves itself over time. The
8:06leases that haven't started yet will
8:08come onto the balance sheet as real
8:10liabilities once the data centers switch
8:13on. That part is just a timing
8:16difference. Meta alone signed $233
8:19billion in new commitments last quarter.
8:2396 billion of it leases that will move
8:26onto the balance sheet as the data
8:28centers come into use. The purchase
8:31commitments mostly turn into chips and
8:34buildings the companies actually own.
8:37What doesn't tidy itself up so neatly is
8:39the cleverer stuff, the joint ventures
8:42and off-balance sheet vehicles
8:44engineered on purpose to stay off the
8:46books. But even that isn't hidden in the
8:50Enron sense. The details are all there
8:53in the accounts. You just have to go
8:55looking for them. So it isn't Enron like
8:58fraud. It's camouflage, which only
9:00really works on people who aren't paying
9:03much attention. Now, none of this means
9:06that big tech plays it straight. They're
9:09plenty aggressive. They just do it in
9:11plain sight on the front page of the
9:13earnings report under a heading that
9:15says adjusted earnings. If you're
9:18spending billions on data centers and
9:21chips, those assets wear out and need
9:24replacing. But many of these firms would
9:26rather talk to you about EBA, earnings
9:29before interest, tax, depreciation, and
9:32amortization.
9:33Charlie Mer suggested that every time
9:36you read the word Ebbita, you should
9:38replace it in your head with BS
9:40earnings. His point about depreciation
9:43was that it's a kind of reverse float.
9:46You pay the cash upfront for the
9:48equipment and the expense shows up later
9:51as the thing wears out. Leaving it out
9:54is just assuming that physical objects
9:56last forever, which is a lovely thought
9:59and very rarely true. And you can see
10:02the strain in the real cash numbers when
10:05the latest earnings landed. The four
10:08biggest hyperscalers posted their lowest
10:10combined free cash flow in a decade. $7
10:14billion between them and Alphabet went
10:17cash negative for the first time since
10:19it went public. Then there's stock-based
10:22compensation.
10:24Tech companies love paying staff in
10:26stock and then taking that expense
10:29straight back out of the earnings they
10:31show investors on the grounds that it's
10:33non-cash. Osw demodin at NYU has called
10:37adding back stock-based compensation one
10:40of the worst abuses in modern reporting.
10:43His point is that it isn't a non-cash
10:46expense in the way depreciation is. It's
10:49a barter. If a company sold shares on
10:52the market and used the cash to pay
10:54employees, everyone would call that a
10:57cash expense. Handing over the shares
11:00directly instead of selling them and
11:02paying cash doesn't make the cost
11:04disappear. Warren Buffett has been
11:06asking the same question for years. If
11:09options aren't a form of compensation,
11:11what are they? If compensation isn't an
11:14expense, what is it? And if expenses
11:17shouldn't go into the calculation of
11:19earnings, where in the world should they
11:21go? To stop all that stock they're
11:24handing to staff from inflating the
11:26share count, the companies use real cash
11:29to buy their own shares back, which they
11:32present to investors as returning
11:34capital. Really, they're running on an
11:37expensive treadmill just to stay in the
11:40same place. And here's the catch. A
11:43buyback only actually rewards the
11:46remaining shareholders if the shares are
11:48bought cheaply. But a company mopping up
11:52its own stock-based compensation doesn't
11:54get to wait for a good price. It has to
11:57keep buying on a schedule whatever the
11:59shares cost that quarter, which lately
12:02has not been cheap. While tech
12:04executives might be aggressive with
12:06their accounting, the actual cash
12:09they're spending on AI is very real. And
12:12it's the way they're spending it that
12:14has some investors worried. Nvidia is
12:17currently working on a round of AI deals
12:19worth more than $750 billion. It's in
12:23talks to backs stop 250 billion to help
12:26open AAI lease computing power and to
12:29finance another 350 billion of OpenAI's
12:33chip purchases. It threw 5 billion at a
12:36secretive new startup run by former Open
12:39AAI chief scientist Ilia Sutsker. Google
12:43has agreed to backs stop lease payments
12:45for Anthropic, effectively handing it a
12:49$35 billion loan. Soft Bank committed 65
12:54billion to Open AI and took out a $40
12:57billion bridge loan just to finance the
13:00bad. If you draw the diagram of who owns
13:03what, the companies at the center of the
13:05AI boom turn out to be mostly investing
13:08in each other. Now, if you were a car
13:12salesman trying to hit your monthly
13:14quota, it might occur to you that
13:16lending a customer the money to buy a
13:18car from you and then booking that as a
13:21sale is a very effective way to move
13:24inventory, at least until the customer
13:26stops making the payments. And the fear
13:29in the market is that AI has turned into
13:32one enormous version of this, a web of
13:36companies funding their own revenue.
13:38Nvidia's CEO Jensen Wong has called the
13:41suggestion that any of this is circular
13:44ridiculous, which is a strong word to
13:47reach for while backstopping a quarter
13:50of a trillion dollars of purchases of
13:52your own product. But to be fair to him,
13:55he has a point. As the Financial Times
13:58pointed out, this is really just
14:00old-fashioned vendor financing. Telecom
14:03equipment makers and plane makers have
14:06been writing checks to help their
14:07customers buy their products for
14:09decades. And the argument for Nvidia
14:12doing it is just as reasonable. The AI
14:16boom is moving fast enough that a
14:18company like Open AI couldn't raise
14:20enough ordinary debt or equity to build
14:23the computing power it thinks it needs.
14:26So, by stepping in, Nvidia locks in a
14:29customer, make sure its chips actually
14:31get used, and if the bet pays off, ends
14:35up owning a slice of something that
14:37could be worth a fortune. The trouble
14:40with vendor financing is what happens
14:43when it doesn't. Then it's a double
14:45blow. You don't just lose the customer,
14:48you lose the money you lent them to be
14:50your customer. And the credit guarantees
14:53make it worse. If an equity state goes
14:56to zero, that's just money wasted.
14:59Annoying, but survivable.
15:01But a promise to cover a customer's
15:04debts if things go wrong can turn a
15:07valuation problem into a solvency
15:09problem. Right now, Nvidia throws off
15:12something like $200 billion a year in
15:15cash. So, if one or two of these
15:17startups trip, it can take the hit. The
15:21question is what happens as the
15:23guarantees climb into the hundreds of
15:25billions and a company that used to
15:28avoid debt is suddenly standing behind
15:30everyone else's. You don't have to take
15:33my word that this matters. The clearest
15:36sign is in Nvidia's own credit market.
15:39The cost of ensuring its debt against
15:41default just jumped by the most on
15:44record in a single day right as this
15:47round of deals landed. So, the people
15:50whose actual job is to price the risk of
15:53Nvidia not paying its bills had a look
15:56at all of this and got noticeably less
15:58relaxed. Because the real risk was never
16:02just that the AI market turns out
16:04smaller than hoped. It's that the people
16:07buying the chips and the people making
16:09the chips are increasingly the exact
16:11same people. All of this circular
16:14financing is happening because everyone
16:17involved is convinced that the market
16:19for AI is going to be so astronomically
16:21large that whatever they spend today
16:24will look like a rounding error tomorrow
16:27as what Demodin has a name for what
16:30happens next. He and his co-author
16:32Bradford Cornell call it the big market
16:35delusion. The way it works is that a new
16:38technology shows up attached to a
16:41massive potential market. A crowd of
16:44companies crop up to serve it and
16:47investors price each company as if it's
16:50going to be the winner. This is not
16:52about the companies talking themselves
16:55up. It's about the people buying the
16:57shares. Each cluster of investors looks
17:00at their chosen company and sees it as
17:03the obvious future giant. The problem is
17:06that they can't all be right. If you
17:09take these companies and add up what the
17:11market expects each of them to earn, you
17:14get a number bigger than the market
17:16itself. Everyone's been priced to come
17:18in first in a race that can have only
17:21one winner. Which is how a whole market
17:24can be priced for a future that
17:26mathematically can't happen. The story
17:29is doing all the work and nobody's
17:31minding the numbers. So, how big is the
17:35story here? The Economist estimates that
17:38the AI buildout is on track to be the
17:41largest investment surge in history.
17:44Around $900 billion this year alone
17:47being spent on chips, data centers, and
17:50power with more than 400 billion of it
17:52borrowed. And then they calculated what
17:56it would take to pay for all of that.
17:59Their estimate is that the industry
18:01would need to be earning something like$
18:03two and a half trillion dollars a year
18:05in AI revenue which is more than the
18:08entire global technology sector earns
18:11from everything it does today. The
18:14actual figure is not close. Adoption is
18:18real. Around a fifth of American firms
18:20report using AI in some way but a lot of
18:24them are using the free versions.
18:27According to a Bank of England study,
18:29the average American executive spends
18:31about a 100 minutes a week using AI.
18:35That's not a typo. The largest capital
18:38investment in the history of the species
18:40is being justified by an hour and a half
18:43per executive per week. So somewhere
18:45between lunch and the drive home. And
18:48when users do pay, they don't pay much.
18:52The fintech firm RAMP went through
18:55actual company spending and found that
18:57the median firm was spending per
19:00employee per month $1066.
19:05$2.5 trillion a year being spent to
19:08capture $10.66
19:11per employee.
19:13The most damning number that came out of
19:15the Bank of England's research was that
19:17nine out of 10 executives said that AI
19:20had made no difference to their
19:22company's productivity over the past 3
19:24years. When a technology takes over the
19:28economy, people usually tend to notice.
19:31But that's the view from the top. Look
19:34at the other end of the economy and the
19:36picture flips. The people getting real
19:39value out of AI aren't the giants
19:42spending hundreds of billions on it.
19:44They're the small ones. According to a
19:46survey by the payroll firm Gusto, the
19:49share of new business founders who used
19:52AI to get started double to 60% in two
19:56years. They are not using it to cure a
20:00disease or replace a department, but to
20:02build a website, handle the local
20:04paperwork and do the things that used to
20:07mean hiring someone. Now, some of this
20:10new business activity is people
20:12incorporating their hobbies. And a
20:14shrinking share of these firms will ever
20:17employ anyone but the founder. So, let's
20:20not oversell it. But the clearest
20:22realworld win for AI so far isn't the
20:26company burning billions on it. It's the
20:28person starting a one-man business
20:30paying about $20 a month. When a market
20:34gets priced as optimistically as AI has
20:37been, the people whose job is to sound a
20:40note of caution sometimes decide to do
20:42the opposite. Take SpaceX. When it went
20:46public in June, it wasn't shy about the
20:49AI framing. its own prospectus claimed a
20:52total addressable market of 28.5
20:55trillion dollars and of that 26.5
20:59trillion or 93% of it was attributed to
21:04AI or Grock which leaves about 2
21:07trillion for everything else the rockets
21:10the launches the satellites the global
21:13broadband network Twitter the actual
21:16space company is the rounding error at
21:19the bottom
21:20That was the case for pricing the shares
21:23at $135.
21:25You'd think that that would be ambitious
21:27enough for anyone, but within weeks, a
21:30Wall Street analyst put a target of $800
21:33a share on it, which would value the
21:36company north of $10 trillion
21:39on a business that did under $19 billion
21:43of revenue last year. Now, you might
21:46wonder why an analyst would look at a
21:48company losing money on 19 billion in
21:51revenue and decide it's worth $10
21:53trillion. As it happens, there is a
21:56reason, and it's a good one. The reason
21:59is what SpaceX is about to do next. One
22:03analyst went through the prospectus and
22:05added up the spending the company is
22:07committed to commitments that the filing
22:10discloses, but never totals in one
22:13place. and they got to something like
22:16$235 billion by 2030. The IPO funds
22:21raised by SpaceX covered only a slice of
22:24that cash requirement. A gap of around
22:28170 billion still needs to be filled by
22:30SpaceX, issuing more stock and more debt
22:34again and again for years, which is a
22:37great deal of underwriting business for
22:40Wall Street. And here's the thing about
22:42those targets. According to Fortune,
22:45analysts at 18 of the banks that
22:48underwrote the IPO put out their price
22:50targets at almost the exact same time
22:54around 25 days after the stock started
22:56trading, which is when the rules let
22:59them start talking. The notes were, in
23:02Fortune's words, almost uniformly
23:04bullish. Morgan Stanley called SpaceX
23:08AI's final frontier. Bank of America
23:11said it was paving the superighway to
23:14the stars. Raymond James compared it to
23:17the invention of electricity, the
23:19railroads, and the internet. These are
23:23supposed to be equity research notes.
23:25Out of 30 odd analysts covering the
23:28stock, exactly one rated a cell, and
23:32that one works at an independent
23:34research firm that doesn't do any
23:36underwriting business. There was for
23:39about 20 years a rule that made this
23:41sort of arrangement awkward, but the SEC
23:44scrapped that rule last December. It was
23:47called the global research analyst
23:50settlement and it dates to 2003 in the
23:53wreckage of the dotcom bubble. It built
23:56a firewall between the investment
23:58bankers and the research analysts at the
24:01same firms. The idea being to stop an
24:04analyst publicly raiding a stock a buy
24:06to help his bank win a fee while
24:09privately emailing colleagues to call
24:11the same company a pig, which isn't a
24:14hypothetical. In the cases that led to
24:17the settlement, one analyst did exactly
24:20that. Another described the stock he was
24:23recommending as a POS, which I'll let
24:26you expand for yourself. The firewall
24:29was taken seriously. Bankers and
24:32analysts at the same firm weren't
24:34allowed to talk business without a
24:36chaperone on the line. Two of the
24:39highest paid people in Manhattan needing
24:41a babysitter on the phone in case they
24:44said something a bit too useful to each
24:47other. The whole thing was championed by
24:50the New York Attorney General at the
24:52time, Elliot Spitzer, better known as
24:55client number nine of the Emperor's Club
24:57VIP. a man with a wellocumented
25:00understanding of the value of keeping
25:02certain transactions off the books. The
25:06SEC scrapped the settlement last
25:08December, citing the need for lower
25:11compliance friction, which is the
25:14regulatory way of saying that the rule
25:16had become a hassle to enforce. So, they
25:18stopped and it's not an isolated
25:21decision. By almost any measure,
25:24American enforcement of financial crime
25:26has been falling apart for years. White
25:29collar prosecutions have been drifting
25:32down over the last 30 years. They're now
25:34running at about half the level of 20
25:37years ago. According to The Economist,
25:40the Justice Department has cut its
25:42lawyers by a fifth and moved
25:44investigators onto immigration and
25:46drugs. The SEC brought a grand total of
25:5010 enforcement actions against auditors
25:53last year, a fifth of its usual rate,
25:56running down the exact oversight that
25:59was built after Enron. And since most of
26:02these crimes carry a 5-year statute of
26:04limitations, the trick is often just to
26:07keep the plate spinning long enough that
26:09the clock runs out. One study reckons
26:12only about a third of corporate fraud is
26:15ever caught at all.
26:17But here's the turn, and it's really the
26:20whole point of this video. For the
26:22companies we're actually talking about,
26:24none of this matters. The Metas, the
26:26Oracles, the Hypers Scalers, they aren't
26:29committing fraud. They don't need to.
26:32Everything they're doing is legal and
26:34nearly all of it is disclosed, which is
26:37a far more interesting situation than
26:40will they get caught because they won't.
26:42The real question is this. If it's all
26:45sitting there in the open, the adjusted
26:48earnings, the 50 billion in leases, the
26:50stock-based compensation added back,
26:53does it actually work? Does dressing up
26:56numbers that anyone is technically free
26:58to read still fool people? Are investors
27:02really being taken in? It turns out that
27:06this is one of the most heavily studied
27:08questions in all of finance. And the
27:10answer is a deeply unsatisfying
27:13yes a bit. And here's exactly how. Take
27:17as what Demodron who we heard from
27:19earlier. His view is that markets are
27:22roughly efficient over time, but that
27:25presentation still matters because most
27:28investors anchor onto whatever number is
27:30put in front of them. Show them an
27:33adjusted figure and that's the only one
27:35they'll use. His objection to adding
27:38back stock-based compensation is exactly
27:41this. It's a real cost dressed up as a
27:44non-cost and a lot of people simply
27:46accepted at face value. Then the harder
27:50evidence. In 1996, an accounting
27:53professor named Richard Sloan published
27:56one of the most famous papers in the
27:58field. He split company earnings into
28:01two parts. The cash the business
28:03actually took in and the acrruals. the
28:06softer judgment-based part that depends
28:09on management's assumptions. And he
28:11found something the market apparently
28:14hadn't. The acrual part is much less
28:17reliable than the cash part. It tends
28:20not to last. Companies whose profits
28:22leaned on acrruels went on to
28:25disappoint. Companies whose profits were
28:27backed by real cash went on to do
28:29better. But the market was treating both
28:33kinds of profit as if they were
28:35identical, which meant that you could
28:37earn excess returns for years simply by
28:41betting that it would eventually notice
28:43the difference. In plain terms, cleaner
28:46earnings beat dressed up earnings. The
28:49polish doesn't hold. So why doesn't the
28:53market just spot this immediately?
28:55That's the second idea. A professor
28:58named Robert Bloomfield has a nice
29:00explanation that he calls the incomplete
29:03revelation hypothesis. In theory, a
29:07market instantly prices in all public
29:09information. But Bloomfield's point is
29:12that public and usable are not the same
29:15thing. The number you need is
29:17technically out there. It's on page 83
29:21of a 200page filing split across four
29:24footnotes in a form you have to
29:26reassemble yourself. And extracting it
29:29cost time, effort, and attention which
29:32aren't free. So the harder a fact is to
29:35dig out, the less completely it shows up
29:38in the price. It isn't that the
29:40information is hidden, it's that reading
29:43it is annoying and most people don't
29:45bother, which is the entire game. The
29:48debt isn't hidden. It's just filed
29:51somewhere tedious enough that you won't
29:53look. At which point, you might
29:56reasonably ask, "Fine. But if there's
29:58money to be made reading the footnotes,
30:00why don't the professionals just do it?
30:03Read the filing, short the overpriced
30:05stock, and collect." And that's the
30:07third idea, the limits of arbitrage, a
30:10field that the economists Mitchell and
30:12Pulino did much of the foundational work
30:15on. The problem with being right about a
30:18footnote is that being right isn't
30:20enough. You also have to stay solvent
30:23long enough for everyone else to catch
30:25up. If you short a beloved narrative
30:28stock because you found an ugly
30:30commitment buried in the account and a
30:32few million people buy it anyway because
30:35they like the founder, that stock can
30:37keep climbing for a very long time. and
30:40your short position can wipe you out
30:43well before the market ever gets round
30:45to caring. The smart money is aware of
30:48this. So, a lot of the time it simply
30:51doesn't bother getting involved in hype
30:53stocks at all. The mispricing survives
30:57not because nobody can see it, but
30:59because the people who can see it can't
31:01afford to bet against the people who
31:03can't, which pulls the whole thing
31:06together. The people not reading the
31:09accounts outnumber the people who are.
31:12The people who are reading them can't
31:14move the price on their own. And so a
31:17company that buries an inconvenient
31:19number in a footnote is making a
31:21perfectly rational bet. That the crowd
31:24won't read it and the professionals who
31:27do won't be able to do much about it.
31:30It's disclosed. It's legal and it works
31:33often enough to maybe be worth doing. So
31:36to bring this all together, the current
31:39panic over hidden tech debt isn't the
31:41discovery of the next Enron. It's people
31:44finally reading the footnotes and
31:46realizing how much money is actually
31:48being spent. As what the motor puts it,
31:51evaluation is a story disciplined by
31:54numbers. Right now, AI is a very
31:58expensive story told by companies
32:00lending each other the money to buy
32:02their own products and reported through
32:05adjusted figures that leave out the most
32:08expensive parts. The debt isn't hidden.
32:12It's just filed somewhere tedious enough
32:14that most people won't look. The problem
32:17isn't that AI is a fraud. It's obviously
32:21useful and the businesses getting the
32:23most out of it seem to be the small
32:25ones, the solo founders and the
32:27one-person businesses paying their $20 a
32:30month, which is a wonderful deal if
32:33you're the one renting. It's a rather
32:35worse deal if you're the one who spent
32:37billions building it and is still
32:40waiting for that $20 a month to add up
32:42to the 2.5 trillion a year it would take
32:45to pay the thing off. Until it does, the
32:49companies building all of this will keep
32:51borrowing. They'll keep filing the
32:53commitments in the footnotes and the
32:55market will keep not reading them right
32:57up until one day it decides to. Jaime
33:01Diamond put it well recently. Will AI
33:04pay off? Probably the way the internet
33:07did. Will it pay off the way you expect
33:10on the timeline you expect? Definitely
33:12not. which means that I should probably
33:15get back to the mirror and keep
33:17practicing my shocked face. I have a
33:19feeling I'm going to need it. If you
33:21found this video interesting, you should
33:23watch my video on Venezuela's missing
33:26oil revenues next. Don't forget to check
33:28out our sponsor, Printify, using the
33:31link in the video description. Have a
33:33great day and see you in the next video.
33:36Bye.
33:38>> [music]

About this transcript

This page contains the full transcript of Why Wall Street is Ignoring Big Tech's Debt by Patrick Boyle, generated from the public captions YouTube serves with the video. The transcript has 5,015 words across 748 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.