Latest / Zero Shot Podcast DEMO / Navigating the AI Investment Bubble: Risks and Realities

Transcript

Navigating the AI Investment Bubble: Risks and Realities · Zero Shot Podcast DEMO

• 0:03
  speakers: Last month, the Bank for International Settlements, which is basically the central bank for central banks, published a study of more than 1 ,000 AI companies.
• 0:11
  The study focuses on one question.
• 0:14
  Using new data on AI firms, investment deals, and supply chain linkages, the study examined investment patterns among the largest AI firms
• 0:21
  globally, focusing on circular investment relationships.
• 0:25
  What they found is that between 2021 and 2025, 55 % of the investment money flowing into AI companies came from other
• 0:33
  AI companies.
• 0:34
  So more than half the money that's funding the AI bubble is coming from inside the bubble, and it only gets messier as the checks get bigger.
• 0:41
  Some of these investors are also doing business with the companies they invest in, like a chipmaker putting money into an AI company, and that same AI company turning around
• 0:49
  and buying its chips.
• 0:51
  Deals like that make up only about 1 in 6 of the deals between AI companies, but when you look at the size of those deals, they're nearly half of all the
• 0:58
  money.
• 0:59
  Which means the bigger the check, the more likely the investor is also the supplier or the customer.
• 1:04
  The study calls these relationships circular.
• 1:07
  The Bank for International Settlements flagged the problem in one very polite sentence.
• 1:11
  These circular arrangements tend to be opaque, making monitoring and supervision difficult.
• 1:15
  Which is basically the same as the referees telling you they can't see the field.
• 1:19
  And until recently, that's where this story stopped.
• 1:21
  Everybody knew about the circular financing problem in AI, but it stayed mostly a so -what type of hypothetical.
• 1:28
  The kind of thing that gets a thoughtful nod at a conference right before everyone heads to lunch.
• 1:34
  But recently, that's changed.
• 1:35
  Now we're seeing people starting to react to the risks in the AI bubble.
• 1:39
  And you can actually watch them react, because there's one corner of Wall Street where fear has a price tag.
• 1:44
  That corner trades a product called a credit default swap, which is insurance on a company's debt.
• 1:49
  And if that sounds familiar, it's because this is the instrument Michael Burry used in the Big Short to bet against the housing market.
• 1:56
  It works like any other insurance.
• 1:58
  You pay a premium, and if the company stops paying its lenders, the insurance pays you.
• 2:02
  Meaning the more worried the market gets, the more that premium costs.
• 2:06
  Most of the time, nobody talks about this market.
• 2:08
  It's a quiet corner of finance that only makes noise when something's about to break, which means when it does start making noise, it's worth paying attention to.
• 2:15
  And as you can see from this chart, the CDS market is waking up.
• 2:19
  This shows the average amount of default insurance on AI companies traded each month over the last five years.
• 2:25
  Since the summer, the people who lent the AI bubble its money have been buying insurance against it at a record pace.
• 2:31
  In July, about $2 billion a week of default insurance was trading on the seven big AI linked names.
• 2:37
  By September, it was over $9 billion a week, more than four times the volume in just two months.
• 2:42
  And the prices followed the volume.
• 2:44
  A few weeks ago, insuring Meta's debt hit a record of about 119 basis points.
• 2:50
  That means investors are paying more for protection against a potential debt default.
• 2:54
  And investors don't pay up for protection unless they're starting to get worried.
• 2:58
  Then NVIDIA, Broadcom, Alphabet, Oracle, and others all hit record highs as well.
• 3:03
  And at the riskier end of AI, the numbers stop looking like insurance and start looking like a prediction.
• 3:08
  But before we get into that, let's take a quick pause to talk about something else.
• 3:12
  According to Microsoft, the average worker gets 117 emails a day.
• 3:17
  It starts before breakfast.
• 3:18
  Of the people online at 6 a .m., 40 % are already in their inbox.
• 3:22
  Then the workday gets in the way, and the replies pile up.
• 3:25
  By 10 p .m., 29 % of active workers are back in their inboxes.
• 3:29
  That's hours of your week disappearing into a keyboard.
• 3:32
  Which brings us to the sponsor of today's video, Whisperflow.
• 3:36
  Whisperflow is a voice -to -text tool that actually works.
• 3:38
  Here's an email that would usually take me about two and a half minutes to type.
• 3:42
  With Whisperflow, the same email took me 30 seconds.
• 3:45
  And there's nothing to fix.
• 3:47
  Whisperflow isn't just for email, either.
• 3:49
  Here's the same text spoken into my phone twice.
• 3:51
  Whisperflow on the left, built -in voice on the right.
• 3:54
  One of these I can send, the other one I have to fix first.
• 3:57
  Whisperflow turns your speech into clear, polished writing across any app and any device.
• 4:02
  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.
• 4:10
  Thank you, Whisperflow, for sponsoring this video.
• 4:12
  Now back to what the market's showing us with the AI bubble.
• 4:15
  The five -year CDS spreads for CoreWeave climbed past 1 ,100 basis points, while the yield on the company's 2031 bonds rose to
• 4:23
  16%.
• 4:23
  At those levels, the market is implying about a 64 % risk of default for CoreWeave over the next five years.
• 4:30
  That's more than a coin flip's chance.
• 4:32
  The market is pricing in a serious probability that CoreWeave won't be able to pay back its debt.
• 4:37
  And this is a company the AI bubble treats as critical infrastructure.
• 4:41
  What's weird is that nothing big in the AI bubble has actually defaulted yet.
• 4:45
  There isn't a single massive bankruptcy to point to.
• 4:48
  But the price of insuring these companies against one is hitting new highs anyway.
• 4:52
  And every insurance contract has two sides.
• 4:55
  Someone collecting the premium and someone paying it.
• 4:57
  So who's actually moving this market?
• 4:59
  Well, you can't buy a credit default swap through your brokerage, and there's no buy button for it on Robinhood.
• 5:05
  In the U .S., this market is effectively closed to everyday investors by law.
• 5:09
  To trade over -the -counter swaps, you need more than $10 million in investments just to qualify.
• 5:14
  So the market is almost entirely made up of banks, hedge funds, and institutional money managers, trading hundreds of billions of dollars a day.
• 5:22
  Part of the reason why comes down to one simple rule.
• 5:25
  Banks cap how much they'll lend to any one industry.
• 5:28
  And last year, the Chicago Fed found that the average large American bank had promised loans to AI -adjacent industries worth about a quarter of their Tier
• 5:36
  1 capital.
• 5:36
  That's the bank's core capital, which is supposed to act as a safety net.
• 5:40
  It's there to make sure a bank can absorb large losses without threatening the stability of the institution itself.
• 5:46
  Loan promises are now worth a quarter of that safety net, all tied to a single theme.
• 5:51
  And when these banks started pressing against their own limits, they started looking for somebody else to hold the risk.
• 5:57
  In May, J .P.
• 5:58
  Morgan, Morgan Stanley, and one of the largest banks in Japan started looking for ways to offload their rapidly growing pile of data center debt.
• 6:05
  That same month, a survey of credit insurers who absorb risk from banks found several nearing their sector limits, or, quote, are already full
• 6:13
  on data center exposure.
• 6:15
  By September, HSBC was working on a deal to hedge the default risk on more than a billion dollars in data center loans by transferring it
• 6:23
  to outside investors.
• 6:24
  Then, last week, the largest AI chip financing deal on record was signed, $63 billion raised to put chips in front of
• 6:32
  Anthropic, the AI lab behind the chatbot Claude.
• 6:35
  And the structure behind that deal is the strangest part.
• 6:38
  The money doesn't even go to Anthropic.
• 6:40
  It goes to a shell company that buys the chips, and Anthropic leases them, which is a key detail we'll come back to.
• 6:46
  This was also a deal that almost didn't even close.
• 6:49
  Some of the banks underwriting it felt they weren't being paid enough for the risk, and considered walking away.
• 6:54
  Broadcom's CEO spent the weekend before the announcement personally talking them off the ledge, but only in exchange for higher interest rates and more protection
• 7:02
  around Broadcom's guarantees.
• 7:03
  And two days after that deal closed, Bloomberg reported that Barclays is looking for default protection on more than $10 billion of AI
• 7:11
  infrastructure debt.
• 7:16
  Banks are still lending to AI companies but are offloading the risk to insurers and other investors, reflecting a precarious balance in the financial system reminiscent of past bubbles.
• 7:25
  The stated reason is to create capacity for more lending.
• 7:29
  So the banks aren't pulling back from AI.
• 7:31
  They're still writing the loans and closing the deals.
• 7:33
  They just don't want to be the ones holding all the risk when they do.
• 7:36
  And we've seen a version of this from Wall Street before.
• 7:39
  It's the same playbook from the last Great Bubble, and we all know how that one ended.
• 7:43
  In 2005, Michael Burry started buying credit default swaps on mortgage bonds.
• 7:48
  He later said he was astounded at how willing the big investment houses were to sell him swaps on highly risky packages of mortgages.
• 7:55
  But a year and a half later, Wall Street caught on and started to agree with him.
• 7:59
  And we know exactly what that looked like from the inside, because a Senate investigation later reconstructed everything using the bank's own emails.
• 8:07
  It started on December 14, 2006.
• 8:09
  Goldman Sachs' mortgage desk had just lost money 10 days in a row.
• 8:14
  That was enough for the chief financial officer, David Vinayar, to pull the mortgage team into his conference room.
• 8:19
  His instruction was simple.
• 8:21
  Get closer to home.
• 8:22
  By closer to home, Vinayar meant for the mortgage department to assume a more neutral risk position.
• 8:27
  The next day, he summed up the meeting in an email to Tom Montag, one of the firm's most senior executives.
• 8:33
  On Goldman's big mortgage bet, he wrote, The position is reasonably sensible, but is just too big.
• 8:38
  And
• 8:48
  Goldman Sachs has not only insured its holdings but has also been actively involved in the risky mortgage market, reflecting a pattern seen in previous financial crises.
• 8:57
  But Goldman didn't stop at just insuring its holdings.
• 9:02
  By the end of the first quarter of 2007, Goldman's mortgage department had swung from a $6 billion net long position in December
• 9:09
  2006 to a $10 billion net short position in late February 2007, a $16 billion reversal.
• 9:17
  A senior Goldman executive later described a $3 billion net short position in subprime mortgage -backed securities as, huge and
• 9:25
  outsized.
• 9:26
  Goldman's net short position in February 2007 was $10 billion, more than triple that size.
• 9:32
  That summer, as the market started collapsing, Vinayar got an update on the mortgage desk numbers.
• 9:37
  His response?
• 9:38
  Tells you what might be happening to people who don't have the big short.
• 9:41
  By the end of 2007, those bets against the mortgage market had brought in about $3 .7 billion.
• 9:48
  The positions were so large and risky that the mortgage department repeatedly breached its risk limits, and Goldman's senior management responded by repeatedly giving
• 9:56
  the department new, higher temporary risk limits to accommodate its trading.
• 10:00
  But here's the part that matters for where we are today with AI.
• 10:03
  While Goldman was betting against mortgages, they were still selling them.
• 10:07
  In 2006 and 2007 alone, Goldman built and sold about $100 billion of mortgage deals.
• 10:13
  One of them was a billion -dollar deal called Timberwolf, a hybrid cash and synthetic CDO squared.
• 10:19
  It closed in March 2007, around the same time Goldman was flipping its bet to the downside, and within five months, the deal had lost 80
• 10:27
  % of its value.
• 10:28
  And Tom Montag, the same executive who received Vinayar's email in December, later gave his opinion on the deal in an email that was read aloud
• 10:36
  at a Senate hearing.
• 10:37
  Boy, that Timberwolf was one shitty deal.
• 10:40
  So that's what the inside looked like last time around.
• 10:43
  The firms closest to the risk saw it first.
• 10:45
  They didn't announce it.
• 10:46
  They didn't stop.
• 10:47
  They sold what they could, insured what they couldn't, and kept the deals moving.
• 10:51
  In July 2007, Citigroup's CEO, Chuck Prince, described this all perfectly when he was asked why his bank was still making risky
• 10:59
  loans.
• 10:59
  His answer became one of the most famous lines in finance.
• 11:03
  When the music stops, in terms of liquidity, things will be complicated.
• 11:07
  But as long as the music is playing, you've got to get up and dance.
• 11:10
  We're still dancing.
• 11:11
  That music he's talking about is liquidity.
• 11:13
  And 19 years later, the same music is back.
• 11:16
  Only this time, it didn't come from the head of a bank.
• 11:18
  It came from the richest man on earth.
• 11:20
  A few weeks ago, Elon Musk posted on X and said, Debt markets have almost entirely shut off access to money for any company that doesn't have real
• 11:28
  revenue outside of AI data centers.
• 11:30
  Then he explained why his own company was fine.
• 11:32
  He said, SpaceX just closed $40 billion of debt for GPUs, but were investment grade and have Starlink and launch revenue.
• 11:40
  Then came the prediction.
• 11:41
  The more than 200 neoclouds, which are cloud companies that rent out AI computing power, Musk says they're headed for, quote,
• 11:50
  And anyone who's ever owned a fish tank knows how guppy deaths go.
• 11:53
  There's no funeral and no announcement.
• 11:56
  One morning, the tank just has fewer fish in it, and the big ones look well fed.
• 12:00
  Musk never used Chuck Prince's words specifically, but it's exactly what he's describing.
• 12:04
  A debt market slowing its lending on the AI story and starting to lend on everything else a borrower has.
• 12:10
  We're three years into the most hyped technology bubble in history, and the survivors are being selected by their non -AI income.
• 12:16
  That's what it looks like when market liquidity starts to go.
• 12:20
  Because market liquidity isn't a number on anyone's balance sheet.
• 12:23
  It's an assumption.
• 12:24
  The assumption that when you need to borrow or you need to sell, somebody will be standing on the other side at close to yesterday's price.
• 12:31
  And market liquidity is there right up until everybody checks for it at once.
• 12:35
  Which leaves one question.
• 12:36
  If the debt market really has shut its doors on companies that live on AI alone, how did the largest AI chip financing deal in history close last
• 12:44
  week?
• 12:45
  Anthropic is one of the most important AI companies on the planet.
• 12:48
  But from a lender's perspective, they're still exactly the kind of borrower Musk is talking about.
• 12:53
  They're only five years old.
• 12:55
  Nearly every dollar they earn comes from AI, the one thing lenders have gotten nervous about.
• 12:59
  And they have committed to more than $500 billion in future infrastructure spending.
• 13:04
  Which means a company like Anthropic walking into the debt markets alone for $63 billion should get an ugly rate attached to that debt.
• 13:12
  If they even get an answer from lenders at all.
• 13:15
  Except Anthropic didn't walk into the debt markets alone.
• 13:18
  They tagged in Broadcom with them.
• 13:19
  It's the same reason a landlord will rent a 22 -year -old their first apartment if a parent co -signs the lease.
• 13:25
  The landlord isn't approving the kid.
• 13:27
  He's approving the parent's credit score.
• 13:33
  The American economy.
• 13:34
  Last year, the research firm Renaissance Macro found that, for the first time ever, AI investment added more to U .S.
• 13:41
  GDP growth than all of American consumer spending combined.
• 13:44
  The American economy is increasingly reliant on AI investments, which have added more to GDP growth than all consumer spending combined, raising concerns about economic stability.
• 13:55
  And J .P.
• 13:55
  Morgan found that, since ChatGPT launched in 2022, just 42 AI -linked companies produced about 65 to 75
• 14:03
  % of the market's returns, profits, and capital spending.
• 14:06
  Their conclusion was, The generative AI investment theme has smothered the rest of the U .S.
• 14:12
  equity market.
• 14:12
  If you take these 42 AI -linked companies out, the American stock market would have underperformed Europe, Japan, and China.
• 14:20
  So the music hasn't stopped yet.
• 14:22
  None of the big AI borrowers has defaulted.
• 14:24
  The deals keep closing.
• 14:25
  And by every official measure, the dancing continues.
• 14:28
  But you don't want to be the one still on the floor when it ends.
• 14:31
  Four months after Chuck Prince said Citigroup was still dancing in 2007, he was fired.
• 14:35
  His bank went on to take some of the biggest mortgage losses on Wall Street, and they needed one of the largest bailouts in American history to survive.
• 14:42
  $45 billion from taxpayers, plus a government guarantee, on more than $300 billion of their assets.
• 14:49
  So the institutional money isn't calling the top.
• 14:52
  It's doing something quieter.
• 14:53
  The banks are still writing the loans.
• 14:55
  The chipmakers are still shipping the chips.
• 14:57
  The deals keep getting bigger, just like we saw in 2007.
• 15:00
  But the risk is being passed around.
• 15:02
  From the banks, to insurers, and credit default swap sellers, and from the borrowers back onto the chipmakers and big tech, who are now
• 15:10
  guaranteeing other companies' bills just to keep the deals closing.
• 15:14
  We even have the central bank for central banks saying these circular arrangements tend to be opaque, making monitoring and supervision difficult.
• 15:21
  More than half the money flowing into AI companies comes from other AI companies.
• 15:26
  America's economic growth has been leaning on that money, and the people financing it are now quietly buying insurance against it.