Is the AI Bubble About to Pop? (Explained Like You're 5) — Transcript
Full transcript
- 0:00July 2026. Leopold Aschenbrenner, a
- 0:03young former OpenAI researcher, runs a
- 0:05fund that bets billions on AI. He's
- 0:08right. AI keeps growing. And still, in a
- 0:10single month, his fund shrinks from
- 0:12about $45 billion
- 0:14to about $10 billion. How can you be
- 0:16right about the future and still lose
- 0:18that much money? Then, last week, AI and
- 0:21chip stocks lost more than $1 trillion
- 0:24in value in less than a week. So, is the
- 0:27AI bubble about to pop?
- 0:28>> Pop? I use AI every day. It's amazing.
- 0:31>> The internet was amazing, too. Its
- 0:34bubble still popped because bubbles
- 0:36don't pop when the technology stops
- 0:37working. They pop when the money runs
- 0:39out. And the money behind AI is doing
- 0:41strange things. Some of it goes in a
- 0:44circle. And one of the richest companies
- 0:46on Earth just borrowed money for 100
- 0:48years. If it pops, it's not just
- 0:50billionaires who pay. If you save for
- 0:52retirement, it might be you. So, let's
- 0:55follow the money step by step. And by
- 0:57the end, you'll know three warning signs
- 0:59to watch for yourself. First, what is a
- 1:02bubble? Let's make AI tiny. Imagine a
- 1:05small village by a river. One morning,
- 1:07someone finds gold in the water.
- 1:09>> Gold? If I find enough, I'm rich.
- 1:12>> To dig, Mia needs a shovel. And the only
- 1:14one who makes shovels is Hank. Soon, the
- 1:17whole village wants to dig. So, everyone
- 1:19wants a shovel.
- 1:20>> Shovels! Get your shovels! Gold not
- 1:22included.
- 1:23>> So, who gets rich in a gold rush? Take a
- 1:26guess.
- 1:27Often, not the diggers. The ones selling
- 1:29the shovels. Today, AI is the gold. The
- 1:32diggers are companies like Google,
- 1:34Microsoft, Meta, and OpenAI. And the
- 1:36biggest shovel seller is Nvidia, which
- 1:38makes the chips AI runs on. So, when
- 1:41does it become a bubble? The famous
- 1:43investor George Soros said, "Every
- 1:45bubble starts with something real. There
- 1:47really is gold in the river. But then,
- 1:49people make a mistake. They think
- 1:51there's more gold or that it comes
- 1:52faster than it really does. And that
- 1:54mistake is what pushes prices way too
- 1:57high. So, AI is useful doesn't answer
- 1:59the question. The gold can be real and
- 2:02it can still be a bubble. That's exactly
- 2:04what happened with the internet. Let's
- 2:05go back to the late 1990s. Companies
- 2:08laid glass cables all over America
- 2:10because a famous number said internet
- 2:12traffic doubled every 100 days. That's
- 2:14about 12 times more every year. In
- 2:17reality, it doubled about once a year.
- 2:19Sounds like a small mistake, but after 3
- 2:21years, it's the difference between 2,000
- 2:24times more and eight. So, they built way
- 2:27too much. After the crash, only a few
- 2:29percent of those cables were even
- 2:30switched on. Cisco, which sold the
- 2:32shovels of the internet, was briefly the
- 2:34most valuable company in the world in
- 2:36March 2000. Two years later, it had lost
- 2:39about 80% of its value. But the internet
- 2:42was real. It just came slower and paid
- 2:44less than people had bet. And it wasn't
- 2:46the first time. In the 1920s,
- 2:49electricity, cars, and radio were brand
- 2:51new and all of them were real. The stock
- 2:54market still crashed in '29. So, the
- 2:56real question isn't is AI useful? It's
- 2:59do the numbers add up? And to check
- 3:01that, we have to look at how much money
- 3:03is going in. This year, the four biggest
- 3:05diggers, Microsoft, Google, Amazon, and
- 3:07Meta, plan to spend over 700 billion
- 3:10dollars, mostly on AI data centers. A
- 3:13data center is a giant building full of
- 3:15computer chips. It's where AI does its
- 3:17thinking. 700 billion is hard to even
- 3:20imagine. So, picture this. You spend a
- 3:22thousand dollars every single second day
- 3:25and night, no sleep. To spend 700
- 3:27billion, you would need about 22 years.
- 3:30Why so much? Because nobody wants to be
- 3:32too late. Meta's boss, Mark Zuckerberg,
- 3:35said he'd rather waste a couple hundred
- 3:36billion dollars than be late. These are
- 3:38the richest companies on Earth. Surely
- 3:41they can pay that from their own pocket?
- 3:42Mostly, yes. But between April and June,
- 3:45Google spent more on building than its
- 3:47whole business brought in. For the first
- 3:49time since it joined the stock market in
- 3:512004. And when your own money isn't
- 3:53enough, you borrow. So, Google borrowed
- 3:55money from investors. It pays them
- 3:57interest every year and pays the loan
- 3:59back at the end. And the end is 100
- 4:02years from now. 100 years. The last tech
- 4:05company that did that was Motorola in
- 4:071997, just before the internet bubble
- 4:09popped. That doesn't mean it will happen
- 4:11again, but it's a strange repeat, and
- 4:14some borrowing is harder to see. Meta
- 4:17needed a giant data center in Louisiana
- 4:19for around $27 billion.
- 4:21Instead of borrowing that itself, Meta
- 4:23set up a separate company with a
- 4:25partner. That company borrowed the
- 4:26money, built the data center, and rents
- 4:28it to Meta. So, the loan isn't in Meta's
- 4:31own books. But Meta promised to pay
- 4:33extra if the building ends up worth
- 4:35less. In the village, it would sound
- 4:37like this.
- 4:37>> I don't have any debt. I just rent this
- 4:39barn.
- 4:40>> And if the barn loses its value?
- 4:42>> Then I pay. It says so right here.
- 4:44>> That's allowed, and it's common. But it
- 4:46means there's more borrowed money behind
- 4:48AI than you see at first. So, a mountain
- 4:51of money is going in, some of it
- 4:52borrowed. That only works if even more
- 4:55money comes back out. So, does it? Let's
- 4:58go back to the village. Hank wants his
- 5:00shovel sales to look amazing, so he has
- 5:02an idea.
- 5:03>> Mia, I'll give you 100 coins, but you
- 5:05have to buy my shovels with them.
- 5:07>> Deal.
- 5:08>> So, Mia buys 100 coins worth of shovels
- 5:11from Hank with Hank's own money.
- 5:13>> Look, I sold 100 coins of shovels.
- 5:16>> But how much new money came into Hank's
- 5:18shop? Zero. It's his own money walking
- 5:20in a circle. Real life is more
- 5:22complicated, but parts of AI look a bit
- 5:24like this. In 2025, Nvidia announced it
- 5:28would invest up to $100 billion
- 5:30in OpenAI. And OpenAI uses money like
- 5:33that to buy Nvidia chips. Nvidia also
- 5:36bought $2 billion of shares in
- 5:38CoreWeave, a company that rents out AI
- 5:40computers. And CoreWeave buys Nvidia
- 5:43chips. And it's not just them. The BIS,
- 5:46a kind of bank for the world's central
- 5:47banks, found that more than half of the
- 5:49money put into AI companies comes from
- 5:51other AI companies. To be fair, the
- 5:54chips are real and they really work. But
- 5:56the circle makes sales look bigger than
- 5:58the new money coming in. So, who brings
- 6:00new money in? Only the people at the
- 6:02very end. The ones who pay for AI. You,
- 6:05me, and businesses. In the village,
- 6:07that's the gold buyer. He's the only one
- 6:09who pays with money from outside. So,
- 6:11how much are customers paying? OpenAI's
- 6:14sales are close to 50 billion dollars a
- 6:16year. Anthropic's are about 67 billion a
- 6:18year going by its last 3 months. That's
- 6:21growing incredibly fast. But together,
- 6:23it's still far less than the 700 billion
- 6:25being spent this year. Now, a data
- 6:27center lasts for years, so it doesn't
- 6:30have to pay for itself in one. But until
- 6:32the customers catch up, someone has to
- 6:34cover the gap. And more and more, that
- 6:36someone is borrowed money. That's where
- 6:39it gets dangerous. Back in the village,
- 6:41meet Rita. She doesn't dig and she
- 6:43doesn't make shovels. She rents them
- 6:45out.
- 6:45>> Can't afford a shovel? Rent one from me.
- 6:48>> To buy all those shovels, Rita borrowed
- 6:50from the village banker.
- 6:51>> 100 coins. You pay me eight coins of
- 6:53interest every year.
- 6:55>> Easy. Mia rents my shovels for 12 coins
- 6:57a year, fixed for 5 years. You only want
- 7:00eight.
- 7:01>> In real life, Rita is a company called
- 7:03CoreWeave. It has borrowed around
- 7:05[clears throat] 52 billion dollars to
- 7:06buy AI chips and rent them out. Like
- 7:09Mia, its customers rent those chips on
- 7:11long contracts. And like Rita, it pays
- 7:13interest on everything it borrowed.
- 7:15Then, last week, the people who lent
- 7:17CoreWeave money got nervous and started
- 7:19selling those loans to others. Here's
- 7:21why that matters. When people are afraid
- 7:23you can't pay them back, they'll only
- 7:25lend you more if you pay more interest.
- 7:27One of CoreWeave's loans pays about 9.6%
- 7:30On the market, buyers suddenly wanted
- 7:32about 15. Its old loans don't change,
- 7:35but every new loan could get that
- 7:37expensive. And CoreWeave still needs
- 7:39billions more for data centers it
- 7:40already promised. Think of it like rent.
- 7:43It's as if your rent jumped from $960 to
- 7:461,500 a month. Same home, just way more
- 7:49expensive. Its stock fell about 15% in
- 7:51one day. Its boss called that irrational
- 7:54and pointed to contracts worth around a
- 7:56hundred billion dollars. Maybe he's
- 7:58right, but there's a second problem and
- 8:00it's not just Coreweave's. Prices in
- 8:02America went up faster than expected.
- 8:04Oil got expensive. So many now expect
- 8:07America's central bank to raise interest
- 8:09rates. And then everything bought with
- 8:11borrowed money gets more expensive. Back
- 8:13in the village.
- 8:14>> Rita, times have changed. Any new loan
- 8:17costs 15 coins a year.
- 8:18>> 15? But Mia only pays me 12.
- 8:21>> That's the trap. Mia still pays Rita 12
- 8:24coins a year, but every new loan now
- 8:27costs Rita 15. She pays out more than
- 8:29she earns. And remember Leopold from the
- 8:31start? This is exactly what happened to
- 8:33his fund. He was right about AI, but the
- 8:36fund had bought with borrowed money.
- 8:38When prices dipped, the people who lent
- 8:40him money wanted cash, right now. So he
- 8:43had to sell. You can be right about the
- 8:45future and still go broke before it
- 8:47arrives. And that's how bubbles usually
- 8:49pop. Not because the technology stops
- 8:51working. The famous investor Ray Dalio
- 8:53put it simply, "Wealth isn't the same as
- 8:56money. You can't spend wealth. To get
- 8:58money, you have to sell it." So when
- 9:00interest goes up and the lenders want
- 9:02their money back, everyone who borrowed
- 9:04has to sell, all at the same time.
- 9:06Prices fall. That forces even more
- 9:08people to sell. Like dominoes. So is AI
- 9:12doomed? Not so fast. The other side has
- 9:15strong arguments. Back in 2000, the
- 9:17cables lay in the ground and nobody used
- 9:19them. Today, big investors say nobody
- 9:22can get enough AI chips. Even old chips
- 9:24still rent for high prices. In the
- 9:26village, back then, shovels piled up in
- 9:29the shop. Today, people wait in line for
- 9:31them. And the money is real. Anthropic
- 9:33sales grew about 45% in just three
- 9:36months because businesses pay for AI
- 9:38that saves them time. Also, the biggest
- 9:40spenders aren't tiny startups like in
- 9:422000. Microsoft, Google, Amazon, and
- 9:45Meta make huge profits from other
- 9:47things. They won't disappear overnight.
- 9:49And even when a bubble pops, the
- 9:50technology usually wins. Those empty
- 9:53cables from 2000, today they carry your
- 9:56videos. But there's one catch that
- 9:57worries both sides. Even if AI changes
- 10:00the world, AI companies might not keep
- 10:02the money. Think about airplanes. They
- 10:04changed the world, but for decades
- 10:07airlines barely made money. Warren
- 10:09Buffett once joked that added up, they'd
- 10:11earned about zero since the very first
- 10:13flight. Why? Every airline sells the
- 10:15same seat from A to B. So, you pick the
- 10:18cheapest. AI could go the same way.
- 10:20People switch between AI apps all the
- 10:22time. One study found OpenAI's share of
- 10:25AI sold to businesses fell from about
- 10:27half to about a quarter in 2 years.
- 10:29Great for you. Not so great for the
- 10:31companies that spent billions. So, who's
- 10:33right? Ray Dalio, one of the few
- 10:35investors who saw the 2008 crash coming,
- 10:38says he sees clear signs of a bubble.
- 10:40But he also says it's not yes or no,
- 10:43it's how big. And anyone who tells you
- 10:45exactly when it pops is guessing. But if
- 10:47it does pop, remember the question from
- 10:49the start. Who pays? First, the owners
- 10:52of AI companies. And that's not just
- 10:54billionaires. Nvidia alone is about 8%
- 10:57of America's 500 biggest companies all
- 11:00added together. So, if your retirement
- 11:02savings follow those 500 companies, like
- 11:04many do, you probably own a small piece
- 11:06of the AI bet, often without knowing it.
- 11:09Second, the people who lent the money.
- 11:12Many AI loans are bought by pension
- 11:13funds and insurance companies. And
- 11:15third, workers. Builders, electricians,
- 11:18and engineers. The International
- 11:20Monetary Fund, which watches the world
- 11:21economy, just said the economy is
- 11:23standing on one leg. The AI leg. In one
- 11:26of its what if tests, a sharp AI crash
- 11:30shrinks the whole world economy by about
- 11:321% by 2028. Not a prediction, but it
- 11:35shows how much now depends on AI. So, is
- 11:38AI a bubble? And if it is, will it pop?
- 11:41And how hard? Nobody knows for sure, but
- 11:44you don't have to guess. You can watch
- 11:46three warning signs yourself, and you
- 11:48already know all of them. Sign one, real
- 11:50customers. Do people and businesses pay
- 11:53more for AI every few months? If yes,
- 11:56the money is real and the spending can
- 11:57pay off. If sales stall, or it's mostly
- 12:00the same money going in circles, the AI
- 12:02companies have spent far more than comes
- 12:04back. Sign two, the cost of borrowing.
- 12:07Remember Rita? Watch what lenders ask AI
- 12:10companies to pay. If borrowing keeps
- 12:12getting more expensive, companies with
- 12:13lots of debt get squeezed first. Like
- 12:16Rita, they may have to sell, and prices
- 12:18can fall like dominoes. Sign three,
- 12:20time. How long until a data center earns
- 12:23back what it cost? Some investors say
- 12:25the best ones do it in one to three
- 12:27years. If that keeps getting longer, the
- 12:29money comes back too slowly. Then,
- 12:31spending gets cut, and the shovel
- 12:33sellers, like Nvidia, feel it first.
- 12:35None of these signs tells you the exact
- 12:37day, but if all three turn red at the
- 12:39same time, the bubble gets more
- 12:41dangerous. If they stay green, AI may
- 12:44grow into its price. So, is it popping?
- 12:47Nobody knows. But Leopold's story shows
- 12:49how it usually ends. Not when the dream
- 12:51dies, when the borrowed money runs out.
- 12:54But the biggest borrower in the world
- 12:55isn't Google or Coreweave. It's a
- 12:58country, and it owes more than 40
- 13:00trillion dollars. Who lent it all that
- 13:02money? What happens when interest goes
- 13:04up? And can it ever be paid back?
- 13:06In the next video, we look at America's
- 13:0840 trillion dollars of debt. See you
- 13:10there.
About this transcript
This page contains the full transcript of Is the AI Bubble About to Pop? (Explained Like You're 5) by PENNY, generated from the public captions YouTube serves with the video. The transcript has 2,457 words across 356 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 choose a caption track available on the video.
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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.