A BOLHA DA COREIA DO SUL ACABOU DE ESTOURAR (explicado) — Transcript
Full transcript
- 0:00What is happening on the other side of
- 0:02the world, in South Korea, could say a
- 0:04lot about the future of artificial
- 0:05intelligence and the global economy.
- 0:08The South Korean stock market is
- 0:09experiencing one of the biggest crashes
- 0:10in its history. In June, it hit an
- 0:12all-time high, and since then, it has
- 0:14fallen over 40%. There have been so many
- 0:17days of panic that the Korean exchange
- 0:19has had to trigger circuit breakers more
- 0:22than 10 times this year alone.
- 0:24Can you imagine experiencing what we
- 0:26went through in October, but 10 times
- 0:27over?
- 0:28And the headline going around the world
- 0:30is that the Korean bubble has burst. But
- 0:32wait a second, what bubble?
- 0:33Did you know that the world's entire
- 0:35artificial intelligence infrastructure
- 0:37depends on a type of memory that comes
- 0:38primarily from South Korea, and that
- 0:40every time you ask GPT a question,
- 0:41there's a good chance Korean memory is
- 0:43making that answer happen?
- 0:45That out of every 10 dollars the world
- 0:47spends on AI chip memory, more than
- 0:49eight go to two Korean companies, and
- 0:52that one of them, Samsung, just had 80
- 0:54billion dollars in operating profit in a
- 0:56single quarter: more than Nvidia, double
- 0:59Google, and more than double Apple?
- 1:01If the companies that manufacture AI
- 1:03memory have never made so much money,
- 1:05why is their stock market crashing like
- 1:07this? Is this really a bubble? Does this
- 1:10reach American giants, and does it reach
- 1:12your money? That is exactly what I am
- 1:14going to explain to you in this video.
- 1:16What is really happening in Korea, why
- 1:18this memory has become the most
- 1:20sought-after piece of artificial
- 1:21intelligence, and what it all means not
- 1:23just for your money, but for the future
- 1:25of the world economy.
- 1:27And for this, I have divided this video
- 1:29into four pillars. First, I will explain
- 1:30the piece at the center of this story,
- 1:32which is the memory. Second, how it
- 1:34became the most sought-after piece of
- 1:36artificial intelligence and made these
- 1:38companies profit like never before.
- 1:39Third, why the Korean market plummeted
- 1:41even with that. And fourth, what this
- 1:43all means for your money. But before I
- 1:45continue, man, I'm going to ask you
- 1:47something, because this video took a lot
- 1:48of work. My team and I dedicated a lot
- 1:50of time to this video, studying company
- 1:52reports, translating Korean articles,
- 1:54and comparing numbers from different
- 1:56sources. So if you made it this far,
- 1:57leave a like on the video and subscribe
- 1:59to the channel. This really helps
- 2:01YouTube understand that this type of
- 2:02video is worth it.
- 2:04And it helps our team a lot to keep
- 2:06bringing this kind of investigation to
- 2:08you. So let's go. Starting with memory.
- 2:10To help you understand, let's build an
- 2:12analogy together here. Think of a
- 2:14professional kitchen. There's the pantry
- 2:16, where everything is stored. There's
- 2:17the countertop, where only the
- 2:19ingredients for the current recipe are
- 2:21kept, and there's the cook. Every device
- 2:23, from your phone to the largest data
- 2:25center in the world, works this way.
- 2:28The cook is the processor, the chip that
- 2:30does the math; the pantry is what we
- 2:32know as storage, where your photos,
- 2:34files, and apps stay even when the
- 2:36device is off.
- 2:38And in a computer, this is the HDD or
- 2:41SSD. And the countertop is the working
- 2:43memory, the one that appears as RAM on
- 2:45the device's spec sheet. It is much
- 2:47faster, but it clears out when you turn
- 2:49it off. Its technical name is DRAM, and
- 2:51it is the most important part of this
- 2:53video. A model like ChatGPT is made of
- 2:56billions of numbers.
- 2:58To write each word of the response you
- 3:00receive, the chip needs to read a huge
- 3:02portion of those numbers, word by word.
- 3:05In the kitchen analogy, it's as if the
- 3:07recipe had billions of ingredients and
- 3:08the cook needed almost all of them at
- 3:10every step.
- 3:12The larger the countertop, meaning the
- 3:14memory, and the faster the ingredient
- 3:15gets to it, the more the cook can
- 3:17produce.
- 3:18And the AI cook is so fast that often
- 3:20the one setting the kitchen's pace is
- 3:22the countertop.
- 3:24That is why AI doesn't just use any
- 3:26memory. It uses a special type called
- 3:28HBM.
- 3:29The acronym stands for High Bandwidth
- 3:32Memory.
- 3:33In practice, it is the memory that
- 3:35delivers the most data per second. HBM
- 3:37is made of the same DRAM as conventional
- 3:39memory, only it is assembled in a much
- 3:42more difficult way. You take several
- 3:44memory chips, eight, twelve, and then
- 3:45you stack them one on top of the other,
- 3:47like in a building.
- 3:49Then you connect the floors with
- 3:51thousands of tiny holes that go through
- 3:53the chip itself, and you glue this
- 3:55building right next to the AI chip.
- 3:58In the kitchen, that would be equivalent
- 3:59to gluing the countertop to the cook and
- 4:01having it deliver the ingredients
- 4:02through thousands of paths at the same
- 4:04time.
- 4:06This memory was created over 10 years
- 4:08ago for computer graphics cards and was
- 4:10a niche product for a long time, until
- 4:13November 30, 2022, when OpenAI launched
- 4:15ChatGPT, and in two months it reached
- 4:18100 million users.
- 4:21No application in history had grown so
- 4:23fast, and each person asking a question
- 4:25meant a chip in a data center reading
- 4:26billions of numbers in memory to be able
- 4:28to answer.
- 4:30That year, Nvidia's main AI chip, the
- 4:33H100, came with 80 GB of HBM.
- 4:36Two years later, the next version
- 4:39already had 141, then 192, and the chip
- 4:41Nvidia just started shipping now carries
- 4:44288 GB.
- 4:46In 4 years, the memory inside each chip
- 4:48has more than tripled, and it wasn't
- 4:49just the chip that grew.
- 4:51The way of using AI has also changed. In
- 4:53the beginning, you asked a question and
- 4:55received a short answer.
- 4:57Today, AI reads entire documents, keeps
- 4:58long conversations, and instead of just
- 5:00answering you, it often thinks and
- 5:01writes out an entire reasoning process
- 5:03you don't even see.
- 5:04And all of this needs to stay in memory
- 5:06while it works. With more people using
- 5:08larger models and each answer demanding
- 5:10more. AI has become the biggest consumer
- 5:12of memory the world has ever seen. In
- 5:142025, the world spent about 35 billion
- 5:17dollars just on HBM, and one HBM
- 5:19manufacturer projects that this figure
- 5:22will reach nearly 100 billion dollars
- 5:25per year by 2028. And this is the memory
- 5:27that I mentioned at the beginning of the
- 5:30video comes mainly from Korea. In the
- 5:32entire world, only three companies know
- 5:33how to manufacture HBM on a large scale
- 5:35. SK Hynix, Samsung, and the American
- 5:38company Micron. And combining the two
- 5:41Korean companies, SK Hynix and Samsung,
- 5:43we are talking about two firms
- 5:45responsible for 80%of the HBM memory in
- 5:47the market. This shows up in the numbers
- 5:49for the entire country. In September,
- 5:51South Korea exported 60 billion dollars
- 5:53in chips alone in a single month, and
- 5:55this is more than triple what it had
- 5:57exported in the same month last year and
- 5:59half of everything the country sold to
- 6:00the world. To give you an idea, that is
- 6:03more than double everything Brazil
- 6:05exports in a month, adding up soy, oil,
- 6:07iron ore, meat, everything. But this
- 6:10jump in export numbers didn't just come
- 6:12from selling more chips; it came mainly
- 6:14because the chips became much more
- 6:16expensive. And to understand why the
- 6:18price went up so much, you need to know
- 6:19one thing about this market. For decades
- 6:22, memory was treated as if it were a
- 6:24commodity, like soy or iron ore. The
- 6:26brand mattered very little. A memory
- 6:28chip is very similar to gold, and what
- 6:30defines the price is supply and demand.
- 6:32That is why the sector has always lived
- 6:34in cycles. When prices rise, everyone
- 6:36builds new factories, the market gets
- 6:37flooded with memory, prices crash, and
- 6:39companies start losing money. Then they
- 6:41cut production, and after a while,
- 6:43memory becomes scarce again.
- 6:45The last bottom was in 2023, and in just
- 6:48the first nine months of that year,
- 6:50Samsung and SK Hynix combined lost over
- 6:5221 trillion won on chips, about 16
- 6:54billion dollars in losses.
- 6:57So, the manufacturers cut production to
- 6:59try to keep prices up. Notice the timing
- 7:01, then. GPT had just been released. The
- 7:04biggest wave of demand for memory in
- 7:06history arrived just when the sector was
- 7:08losing money and producing less.
- 7:10And with memory, you don't fix that
- 7:12overnight. A new factory costs tens of
- 7:14billions of dollars and takes years to
- 7:16finish. To make matters worse, HBM comes
- 7:19from the same factories as standard
- 7:21memory and uses about three times the
- 7:23capacity to deliver the same amount.
- 7:26So, the more manufacturers shifted
- 7:28production to AI, the less there was for
- 7:30everything else: phones, computers, and
- 7:32regular servers.
- 7:33The result was the price of memory
- 7:35skyrocketing. According to this second
- 7:37site, in just the second quarter of 2026
- 7:40, the price of DRAM rose 44%in three
- 7:42months. And storage, the chip in your
- 7:45phone and SSD, went up 53%.
- 7:48In other words, the world is paying much
- 7:50more for memory because there isn't
- 7:51enough for everyone who wants to buy it
- 7:53.
- 7:54In September alone, the global
- 7:56semiconductor industry—including memory
- 7:58, processors, and sensors—sold over 170
- 8:00billion dollars, nearly two and a half
- 8:02times what it sold a year earlier.
- 8:04And by August, total sales for the year
- 8:06had already surpassed 1 trillion dollars
- 8:08. And who is on the other side of the
- 8:09counter? Well, the same Korean companies
- 8:10. Because Samsung and SK Hynix don't
- 8:13just make HBM. They are also the largest
- 8:15manufacturers of standard memory, the
- 8:17kind in your phone and your computer.
- 8:19Together, they make nearly two-thirds of
- 8:22all the DRAM on the planet. Meaning they
- 8:24win both ways: with HBM, where the
- 8:26competition is, and with standard memory
- 8:28, which got more expensive precisely
- 8:30because of it. And so, the numbers for
- 8:32these companies changed completely. Do
- 8:34you remember the losses from 2023?
- 8:36In the third quarter of 2026, SK Hynix
- 8:39reported 104.6 trillion won in revenue.
- 8:42The first time it surpassed 100 trillion
- 8:44in a single quarter. And this is more
- 8:45than four times what they had earned a
- 8:47year before. And for every 100 won that
- 8:50came in, about 80 turned into operating
- 8:52profit. The highest profit in the
- 8:53company's history.
- 8:55To give you a benchmark, at Apple, one
- 8:56of the world's most profitable companies
- 8:58, that number is around 35.
- 9:00At Nvidia, it's 65. And Samsung went
- 9:02down the same path.
- 9:03In the third quarter, it hit its record
- 9:06profit, that 80 billion dollars I
- 9:07mentioned at the start, more than any
- 9:09major American company last quarter.
- 9:12In other words, two companies that were
- 9:13losing money on chips three years ago
- 9:15have become profit machines. At Samsung
- 9:17, the chip division had a margin of
- 9:19about 72%, and at SK Hynix, around 80%.
- 9:22Both of them even higher than Nvidia.
- 9:24And market projections suggest that in
- 9:262026, Samsung will end the year as the
- 9:27most profitable company on the planet.
- 9:30In fact, in early July, at an internal
- 9:32meeting with Samsung's chip division
- 9:34employees, one of the division
- 9:36presidents said that this year's profit
- 9:38alone would exceed everything the
- 9:39division had earned in the previous 40
- 9:41years combined.
- 9:43And here is something very bizarre.
- 9:44Everyone knows Samsung for its phones,
- 9:46the great rival to Apple that it is. But
- 9:49in the third quarter, the mobile and
- 9:50electronics division reported a loss.
- 9:52Meanwhile, the chip division brought in
- 9:5575 out of every 100 won Samsung earned,
- 9:58and virtually all of the company's
- 10:01profit—106.5 out of 107.4 trillion won.
- 10:04In practice, Samsung has become a memory
- 10:06company that happens to also sell phones
- 10:08. And that prosperity has even reached
- 10:10those who work there. SK Hynix shares 10
- 10:12%of its profit with employees, and in
- 10:14February, that resulted in an average
- 10:16bonus of 140 million won per person,
- 10:18which is about 100,000 dollars. And the
- 10:20forecast for the next one, in early 2027
- 10:22, is about half a million dollars per
- 10:24employee.
- 10:26If the profit was reaching the employees
- 10:27like that, imagine what was happening to
- 10:29those who held the stock, right?
- 10:31In 2025, Samsung stock rose 127%, and in
- 10:34May 2026, for the first time in history
- 10:36, Samsung became worth more than 1
- 10:39trillion dollars.
- 10:40And a few weeks later, it became the
- 10:42first Korean company worth 2 quadrillion
- 10:45won.
- 10:46SK Hynix went even further. At the end
- 10:48of 2025, it was worth about 300 billion
- 10:50dollars. By June 2026, it surpassed 1.3
- 10:52trillion, more than quadrupling in less
- 10:55than six months. A company that almost
- 10:57went bankrupt in the early 2000s
- 10:59overtook Samsung to become the most
- 11:01valuable in Korea. A position that
- 11:03belonged to Samsung for over 25 years.
- 11:05Two Korean companies each worth over 1
- 11:08trillion dollars. And even after all
- 11:11that growth, they still looked cheap.
- 11:14There is a number every investor looks
- 11:15at to know if a stock is expensive or
- 11:16cheap, which is the P/E ratio, the stock
- 11:18price divided by the company's earnings
- 11:19.
- 11:20In plain English, how many years of
- 11:22earnings would you need to recover what
- 11:24you paid for the stock?
- 11:26Tech companies usually have a very high
- 11:28number, as the market pays a premium
- 11:30because they grow fast.
- 11:32Nvidia, for example, in July was trading
- 11:34at about 22 times the earnings expected
- 11:36for the next 12 months. Now look at
- 11:38memory chips.
- 11:40In May, SK Hynix was trading at around
- 11:425.6 times expected earnings.
- 11:45And as you will see in a moment, after
- 11:46all the decline, that number got even
- 11:47lower.
- 11:49To bring it to our reality, here in
- 11:50Brazil, Weg, one of the companies the
- 11:52market most loves to overpay for, trades
- 11:54at almost 30 times earnings.
- 11:56SK Hynix, with higher margins than
- 11:58Nvidia and profiting like few others in
- 12:00the world, cost less than six.
- 12:02For investors, buying shares of these
- 12:04companies was an obvious choice. But
- 12:06this logic completely changed their
- 12:07stock market. The Korean stock index,
- 12:10the Kospi, is calculated by the market
- 12:12value of the companies. The more a
- 12:13company is worth, the more weight it has
- 12:15in the index.
- 12:17So, the more Samsung and SK Hynix rose,
- 12:19the more they pulled the whole market up
- 12:21with them.
- 12:22At their peak, the two alone generated
- 12:24more than half of the index. This
- 12:26coincided with another movement in the
- 12:27Korean market. For decades, Korean
- 12:29stocks were cheaper than those of
- 12:31similar companies abroad.
- 12:33Investors were wary of the families that
- 12:34control the country's large
- 12:35conglomerates, right? And the market
- 12:37ends up calling this the Korean discount
- 12:39. And the government that took office in
- 12:412025 enacted reforms to give more voice
- 12:44to small shareholders, precisely to end
- 12:46this discount.
- 12:48Even while still a candidate, the
- 12:49president of Korea had promised to take
- 12:51the stock market to 5,000 points, when
- 12:53it was near 2,500 points.
- 12:55It was a promise to double it.
- 12:56Artificial intelligence on one side,
- 12:58reform on the other, Korea became a
- 13:00trendy market. And in 2025 the Korean
- 13:02stock market rose 75%, one of the
- 13:04biggest gains in the world. And in June
- 13:062026 it surpassed 9,100 points. In just
- 13:09over a year, it nearly quadrupled. And
- 13:11the market thought there was much more
- 13:13to come. In May, JP Morgan predicted the
- 13:16Korean stock market would hit 10,000
- 13:17points by the end of the year. Nomura
- 13:19predicted 11,000.
- 13:21But curiously, on the day the stock
- 13:22market closed above 9,000 points for the
- 13:24first time, out of every 10 traded
- 13:26stocks, more than eight fell.
- 13:28The index was rising, but what was
- 13:29really going up were Samsung and SK
- 13:31Hynix. It was a warning sign. The entire
- 13:33market depended on these two companies,
- 13:35but for those looking from the outside,
- 13:36all that appeared was the index hitting
- 13:38record after record. And when the market
- 13:40hits record after record, everyone wants
- 13:42in. And for many Koreans, it was
- 13:44actually the only way out. In Seoul, a
- 13:46house costs almost 14 times what a
- 13:48family earns in an entire year. The
- 13:50president himself even said that real
- 13:52estate used to be the only possible
- 13:54investment and now there was an
- 13:55alternative. Imagine the mindset of a
- 13:57Korean in their 20s: home ownership is
- 13:59out of reach, impossible. And right next
- 14:02door, two of their own country's
- 14:03companies hitting record after record.
- 14:05Man, the choice that made the most sense
- 14:06wasn't a house. And the Koreans went all
- 14:08in. Many people started borrowing money
- 14:10from brokerages to buy stock. It’s a
- 14:13huge thing in Korea.
- 14:15In June, Koreans already owed 38
- 14:17trillion won to brokerages, more than 25
- 14:19billion dollars, and over a quarter of
- 14:21that had come in just six months.
- 14:24And that is what we call leverage.
- 14:26In May, Korea released a new product:
- 14:28funds that double the daily movement of
- 14:31Samsung or SK Hynix.
- 14:33If the stock goes up 5%, the fund goes
- 14:35up 10; if it falls 5%, the fund falls 10
- 14:37. And on the first day, about 3 billion
- 14:39dollars flowed into these funds. In less
- 14:42than a month, it was already over 9
- 14:43billion. But these funds are
- 14:45double-edged swords.
- 14:47To deliver double the performance every
- 14:48day, this fund needs to buy more stock
- 14:49at the end of the day when it rises and
- 14:51sell when it falls.
- 14:52On the way up, this helps push the price
- 14:54higher; on the way down, it pushes it
- 14:56lower.
- 14:57It was a market sort of hanging on two
- 14:58companies, with the average investor
- 15:00increasingly leveraged and in debt,
- 15:01betting everything in the same direction
- 15:02.
- 15:03And it was exactly there, at the peak,
- 15:04that everything began to turn.
- 15:07While the small Korean investor was
- 15:08entering the market, large foreign funds
- 15:10were going the other way.
- 15:12In May alone, at the height of the rally
- 15:14, they sold 44 trillion won in Korean
- 15:16stocks.
- 15:17It is the largest sale in a single month
- 15:18in the country's history. And a large
- 15:20part of those buying these stocks were
- 15:22local investors, often with money
- 15:24borrowed from the brokerage. That
- 15:26borrowed money worked very well as long
- 15:28as the stock market was rising. The
- 15:30problem is when the tide began to turn.
- 15:32And turn it did. On June 22nd, the
- 15:34Korean stock market closed at a record
- 15:379,114 points. The very next day, it
- 15:39plummeted 9.99%. Samsung and SK Hynix
- 15:42both fell more than 12%each. Their worst
- 15:45day since the 2008 crisis. And as shares
- 15:47dropped like that, brokerages began
- 15:49forced liquidations of positions bought
- 15:52on margin.
- 15:53And every forced sale pushed the price
- 15:55down a little further, which forced
- 15:56brokers to sell even more people's
- 15:58positions.
- 15:59And that creates a domino effect. And
- 16:00that is how Korea arrived at the July
- 16:02crash.
- 16:03By the 30th, the market had accumulated
- 16:06a 34%drop for the month, more than in
- 16:08the worst months of the 1997 and 2008
- 16:10crises.
- 16:12In that month alone, brokerages forcibly
- 16:14sold over half a billion dollars in
- 16:15shares from regular investors.
- 16:17People who didn't want to sell but were
- 16:19forced to. And at the end of the month,
- 16:21both Korean exchanges had to trigger
- 16:22circuit breakers at the same time for
- 16:24two days in a row, something that had
- 16:25never happened there before. Now, you
- 16:27know what’s curious? At the beginning of
- 16:29July, while the market was entering one
- 16:30of the worst months in its history,
- 16:32Samsung announced its biggest profit
- 16:33ever. So why were the world’s major
- 16:35funds selling off exactly the companies
- 16:36making the most profit on the planet?
- 16:38And the answer begins with that story
- 16:39about the cycle. Memory has always
- 16:41worked like this. Memory is scarce,
- 16:43prices go up.
- 16:44Manufacturers make a lot of money, build
- 16:45new factories, and a few years later the
- 16:47factories are ready, there’s a surplus,
- 16:49prices collapse, and profits collapse
- 16:50with them.
- 16:51So, for those who have followed the
- 16:53sector for a long time, record memory
- 16:55profits usually appear right near the
- 16:56peak. In other words, when profits are
- 16:58high, it’s time to get out. And that was
- 17:00when the market split in two. On one
- 17:02side, those who believe that this time
- 17:03it’s different. This group has strong
- 17:05arguments. The first is that the memory
- 17:07shortage continues, right? In May, the
- 17:08CEO of Micron said that the company can
- 17:10only deliver between half and two-thirds
- 17:12of what its main clients ask for.
- 17:15Second, TrendForce, which is one of the
- 17:17leading consultancies in the sector,
- 17:18right?
- 17:20Where they said that the shortage should
- 17:22increase in 2027 and the new factories
- 17:23being built should only make a real
- 17:25difference in 2028, and the situation
- 17:27has reached a point where Nvidia itself
- 17:29is considering putting less memory in
- 17:31one of its next chips, right?
- 17:33Because there might not be enough memory
- 17:35to buy.
- 17:36In fact, the president of ADATA, one of
- 17:38the world's largest memory module
- 17:39manufacturers, even said that the entire
- 17:41production from the three major
- 17:43manufacturers is already sold out
- 17:44through 2027.
- 17:46And according to industry sources, some
- 17:48clients are only receiving 60%to 70%of
- 17:49what they ordered. And at the end of
- 17:51October, the Korean companies themselves
- 17:53reinforced this. Samsung said it has
- 17:54already closed its HBM contracts for
- 17:562027 and that production is sold out.
- 17:59And SK Hynix said the memory shortage is
- 18:01expected to last until 2030. The second
- 18:03argument is that AI memory has stopped
- 18:06functioning as a commodity.
- 18:08HBM is custom-made for each client, with
- 18:10a long-term contract.
- 18:12Micron, for example, already has
- 18:14agreements covering about 35%of its
- 18:16sales through 2030, and this is the side
- 18:18where most banks and brokerages stand,
- 18:20right? At the end of September, the
- 18:23brokerage Yuanta set a price target for
- 18:25Samsung at 630,000 won.
- 18:28Mirae Asset, one of the largest in Korea
- 18:30, set SK Hynix at 3.1 million, and in
- 18:32early November, Samsung was trading near
- 18:34200,000 won and SK Hynix near 1.2
- 18:36million.
- 18:38In other words, according to these
- 18:39analysts, these stocks are worth between
- 18:41two and three times today's price. On
- 18:42the other side, there are those who
- 18:44think it's the same old cycle, just
- 18:46bigger.
- 18:47This group also has a strong argument
- 18:48that Peter Lynch, one of the most famous
- 18:50fund managers in history, explained back
- 18:51in the late 80s.
- 18:53In cyclical companies, the P/E ratio is
- 18:55lowest precisely at the peak, when
- 18:56profits are at their maximum. And that
- 18:58five-times P/E ratio that looks like a
- 19:00bargain is only a bargain if the profit
- 19:01stays where it is. So do the math with
- 19:03me. If a stock costs five years of
- 19:05profit and the profit drops by half, it
- 19:06starts to cost 10. If the profit drops
- 19:08by 75%, it starts to cost 20. This is
- 19:11without the stock price changing by a
- 19:12single cent. And the memory industry has
- 19:13already been through this. In 2018, at
- 19:15the peak of the previous cycle, Samsung
- 19:17made record profits. The following year
- 19:19, memory prices plummeted and the
- 19:20company's profit fell by more than half
- 19:22. And this is the side where the big
- 19:24foreign funds are.
- 19:26In 2026, they have already sold more
- 19:28than 200 trillion won in Korean stocks,
- 19:30and almost 90%of that was just Samsung
- 19:32and SK Hynix.
- 19:33And it makes sense. When a big fund sees
- 19:35a stock quadruple in six months in a
- 19:37sector that has always been cyclical, it
- 19:39does what it always does. He pockets
- 19:41some of the profit and reduces the bet
- 19:43size.
- 19:44And it was this profit-taking that
- 19:45triggered the nearly 10%drop on June
- 19:4723rd.
- 19:48But in a market full of leveraged
- 19:50players, a normal sale becomes an
- 19:52avalanche, because since they were
- 19:54launched in May, the funds that double
- 19:56the movement of the two stocks have
- 19:58received 12.5 trillion won from small
- 20:00Korean investors. And between early June
- 20:03and early July, even before the July
- 20:05crash, they had already lost an average
- 20:07of 27%. And it wasn't just the small
- 20:09investor who was leveraged. The most
- 20:11famous case is Situational Awareness,
- 20:13the fund run by Leopold Aschenbrenner,
- 20:16an ex-OpenAI researcher in his twenties
- 20:18who became a market star betting on
- 20:21artificial intelligence. The fund
- 20:23managed over 20 billion dollars, and one
- 20:25of its bets was precisely SK Hynix.
- 20:28When the stocks plummeted in July, the
- 20:30banks that had lent money to the fund
- 20:31demanded more collateral.
- 20:33With no cash to cover it, on July 30th,
- 20:36the fund sold most of its shares to
- 20:38Citadel, another giant in the US market
- 20:40. And the very next day, with that
- 20:42seller out of the way, SK Hynix rose 24
- 20:44%and the Korean stock market 14%in a
- 20:46single session.
- 20:48You can't prove a single fund crashed
- 20:49the market, but you can see the weight
- 20:51that forced selling from leveraged
- 20:52positions can have.
- 20:54And notice the irony.
- 20:55The day after the sale, the stock he was
- 20:58forced to sell rose 24%.
- 21:00His thesis on artificial intelligence
- 21:02might still prove right down the road,
- 21:03but leveraged the way he was, he didn't
- 21:05have time to wait for the market to
- 21:07catch up. And well, after July, the
- 21:09market showed signs it would recover. In
- 21:11early October, the Korean stock market
- 21:13was around 7,000 points and companies
- 21:15kept breaking records. Korea set a
- 21:17record for chip exports in September,
- 21:20Samsung announced record profits again,
- 21:22and SK Hynix soon after. But two more
- 21:25waves of decline followed, one in
- 21:26October and another in early November.
- 21:28And this time, the arguments from the
- 21:30pessimists gained strength. In '26, the
- 21:32most profitable memory for these
- 21:35companies was standard memory, for
- 21:37phones and computers, not AI HBM.
- 21:40The chairman of the SK Group, which owns
- 21:42SK Hynix, actually said that HBM
- 21:44provides a margin of about 60%, while
- 21:46regular memory reached around 80%. And
- 21:48it makes sense. Remember that HBM comes
- 21:50from the same factories as regular
- 21:52memory and uses three times the capacity
- 21:54to deliver the same quantity? To meet AI
- 21:56demand, manufacturers have been
- 21:58dedicating an increasingly larger
- 22:00portion of their factories to HBM. And
- 22:02today, they already occupy about 20%of
- 22:05all DRAM production capacity in the
- 22:07world. Next year, it should reach close
- 22:09to 30%, but regular memory goes into
- 22:12many more places, right?
- 22:14Phones, computers, cars, and even the AI
- 22:16servers themselves, which use a lot of
- 22:18regular memory in addition to HBM.
- 22:21And with fewer factories left for it,
- 22:22the shortage became even greater exactly
- 22:24there. And there is one more difference
- 22:26. HBM is sold in long-term contracts
- 22:28with prices agreed upon with each client
- 22:30for the entire year. Regular memory has
- 22:32its price renegotiated every quarter. So
- 22:34, when the shortage tightened, its price
- 22:36went up immediately. For 2026, the price
- 22:38of HBM went up, and that increase
- 22:40applies to the whole year. However, the
- 22:42price of regular memory rose nearly 100
- 22:43%in just the first three months of the
- 22:45year.
- 22:46In other words, a huge part of the
- 22:48memory companies 'profits came precisely
- 22:49from the most commoditized part of the
- 22:51business, the part that has always lived
- 22:52in cycles.
- 22:53And that is exactly where China appeared
- 22:55.
- 22:56In October, at a conference in Shanghai
- 22:58, the president of CXMT, China's largest
- 23:00DRAM manufacturer, said they would
- 23:02launch a regular server memory by the
- 23:04end of the year with technology that
- 23:06shouldn't reach Samsung before 2028, and
- 23:08CXMT, which had 4%of the DRAM market a
- 23:10year ago, already has 10%.
- 23:13The company hasn't proven it can produce
- 23:15this at scale yet, but the message to
- 23:17the market was clear. There is a large
- 23:18Chinese competitor growing fast exactly
- 23:20in the memory that is currently the most
- 23:21profitable. Three days later, Samsung
- 23:24and SK Hynix fell more than 7%each. And
- 23:27on another front, HBM also took a hit.
- 23:29It depends on tech giants continuing to
- 23:32build AI data centers. And this
- 23:34investment has become gigantic.
- 23:36Microsoft, Google, Amazon, and Meta
- 23:38alone are expected to spend more than
- 23:40700 billion dollars in '26. And Google
- 23:42and Amazon have already warned that they
- 23:44will spend much more in 2027. But this
- 23:46money no longer fits within their cash
- 23:48reserves. In 2024, for every 100 dollars
- 23:50they invested, about nine came from
- 23:53loans. Today that number is 32, and in
- 23:55the last quarter, Google spent more cash
- 23:58than it generated.
- 23:59Meta invested more than its operations
- 24:01brought in, and Amazon borrowed 48
- 24:03billion dollars in three months.
- 24:06And you might think, well, borrowing to
- 24:07grow is normal. The problem is that this
- 24:09borrowing has become much more expensive
- 24:10.
- 24:11The rate the U.S. government pays to
- 24:13borrow money for 10 years, which serves
- 24:15as a benchmark for much of the world's
- 24:17debt, exceeded 5.3%in October, the
- 24:19highest level since 2002.
- 24:21And even after pulling back slightly, it
- 24:24remains near 5%. And behind this is
- 24:26inflation. With the war in Iran driving
- 24:28oil prices up, gasoline in the United
- 24:30States became 34%more expensive in a
- 24:32year. And U.S. inflation reached 3.7%.
- 24:35The Federal Reserve raised interest
- 24:37rates again in September, for the first
- 24:39time since 2023, and the market is
- 24:41already betting on another hike in
- 24:42December. And those who lend to tech
- 24:44giants were already charging more.
- 24:47According to a Reuters survey from July
- 24:48, of the debt securities these companies
- 24:50issued in 2026, nearly nine out of 10
- 24:52were already paying higher interest than
- 24:54when they were sold.
- 24:56In other words, building data centers
- 24:58depends increasingly on borrowed money,
- 25:00just when that money has become more
- 25:02expensive. And now, besides being more
- 25:04expensive, building data centers may
- 25:05become more difficult.
- 25:07In early November, during the U.S.
- 25:09midterm elections, the Democrats won
- 25:11control of Congress.
- 25:13And on the same day, several cities
- 25:15voted on whether they wanted data
- 25:16centers near their homes.
- 25:18There were 41 local votes, and in 40 of
- 25:20them, the side wanting to restrict
- 25:22construction won with 72%of the votes.
- 25:25That same week, the U.S. chip company
- 25:27index fell more than 12%in a single day
- 25:30. And the Korean stock market hit
- 25:31another circuit breaker.
- 25:33And the revenue that should pay this
- 25:35bill also wavered, because on the same
- 25:37day Samsung announced the record, OpenAI
- 25:39, the owner of GPT, warned investors
- 25:41that annual revenue was around 50
- 25:43billion dollars, not the nearly 80
- 25:45billion that had been disclosed the
- 25:47previous month.
- 25:48And in early November, it managed to
- 25:51raise 11 billion dollars from investors
- 25:53, much less than the 30 billion it was
- 25:55seeking. And it's not just OpenAI.
- 25:57Anthropic, owner of Claude, is going
- 25:59public on the U.S. stock exchange. And
- 26:01in the documents it filed with the SEC,
- 26:02it showed that it generated 17 billion
- 26:04dollars between July and September alone
- 26:06, more than 12 times what it generated a
- 26:08year earlier. But the company has
- 26:10already committed to spending around 900
- 26:12billion dollars, almost all of it on
- 26:13computing and infrastructure.
- 26:16And the market is no longer going to pay
- 26:17for this IPO the way the market was
- 26:19expecting back in August, right?
- 26:21There was talk that Anthropic's IPO
- 26:23wanted to value the company at 2
- 26:25trillion dollars, and now the IPO price
- 26:27range points to 1 trillion dollars. And
- 26:29if the bill for artificial intelligence
- 26:31gets more expensive, the revenue takes
- 26:32longer to arrive. Building data centers
- 26:35becomes more difficult, and that is a
- 26:37disaster for HBM memory. So, returning
- 26:39to the title question: has the
- 26:41artificial intelligence bubble burst in
- 26:43Korea? The honest answer is that it's
- 26:45difficult to pin down an answer. On the
- 26:47optimistic side, analysts are moving in
- 26:48the exact opposite direction of the
- 26:50market.
- 26:51In early October, D.A. Davidson, an
- 26:53American brokerage firm, raised Micron's
- 26:55price target from 2,100 to 3,000 dollars
- 26:58. The stock was around 1,000 dollars,
- 27:00and according to this brokerage, it
- 27:01could practically triple.
- 27:03And the Korean brokerages, as I showed
- 27:05you, still see Samsung and SK Hynix
- 27:06being worth two or three times today's
- 27:08price.
- 27:09And as I said, the argument is that this
- 27:11cycle is different. On the pessimistic
- 27:13side, the rise in the price of standard
- 27:15memory is losing steam.
- 27:16It rose almost 100%in the first quarter
- 27:18of the year, about 60%in the second, and
- 27:20about 15%in the third.
- 27:22For the fourth, TrendForce projects
- 27:24between 10 and 15%. It continues to rise
- 27:26, but by less and less. This weighs more
- 27:28than it seems. At Micron, for example,
- 27:31according to analyst estimates, HBM is
- 27:33less than 10%of revenue, and everything
- 27:35else depends on standard memory prices
- 27:37staying high. Now, there is one thing
- 27:39that is very clear. Regardless of who is
- 27:41right, the expectation is that the
- 27:43Korean market will continue to
- 27:45experience volatilities that feel like
- 27:47crises. And after all the tumble in July
- 27:49, Koreans started borrowing money again
- 27:51to buy stocks. In October, debt with
- 27:54brokerages returned to 34 trillion won,
- 27:56and with the new drops, it shrank to
- 27:58about 28 trillion by early November.
- 28:00And many people continue to see the drop
- 28:02as a chance to buy cheap. In early
- 28:04November, while foreigners spent seven
- 28:06straight trading sessions selling, the
- 28:08small Korean investor was buying. And
- 28:10forced liquidations have started to rise
- 28:11again. There was even an emergency
- 28:13government meeting, right? In early
- 28:15November, they relaxed the loan
- 28:16guarantee rule for stock purchases until
- 28:19the end of January, and brokerages
- 28:20started offering to postpone forced
- 28:22sales.
- 28:24And leveraged funds remain in the middle
- 28:25of this mess.
- 28:26Besides Leopold's fund, another AI fund
- 28:29, Vara, which managed about 20 billion
- 28:32dollars, lost 44%in July and another 33
- 28:34%in October.
- 28:36The fund said it remains stable, but
- 28:37these numbers show just how much this
- 28:39market is shaking. In other words, the
- 28:41Korean stock market remains hung up on
- 28:43two companies with many indebted
- 28:44investors and leveraged funds in the
- 28:46middle of it. And with these pieces on
- 28:48the board, any good or bad news turns
- 28:50into a giant move. And the craziest part
- 28:52is knowing that this moment of
- 28:53volatility and uncertainty isn't just in
- 28:55Korea. Abroad, US interest rates started
- 28:57near the highest level in over 20 years
- 28:59, and that weighs on the entire world's
- 29:01economy, including Brazil's. And here in
- 29:03Brazil, we just saw this up close. One
- 29:05of the weeks in October was, well, the
- 29:06worst week for the Brazilian stock
- 29:08market since the start of the pandemic.
- 29:10The Central Bank had been cutting rates
- 29:12, stopped at a Selic of 13.75%, and even
- 29:14increased the inflation projection. And
- 29:16it's in moments like these of great
- 29:17uncertainty that many people end up
- 29:19making the worst decision of the year
- 29:20with their own money. And a good part of
- 29:22these bad decisions stems from the
- 29:24attempt to guess what will happen, where
- 29:25the stock market is headed, how far US
- 29:27rates will go, and how far the dollar
- 29:29will go. But to get through a moment
- 29:30like this, what helps is knowing what to
- 29:32do with your money in any of these
- 29:33scenarios. And when everything becomes
- 29:35uncertain, having a team by your side
- 29:37studying each company and telling you
- 29:39what price is worth paying makes a big
- 29:41difference. And that is why I cannot
- 29:42fail to talk to you about Finclass.
- 29:44Inside, you have access to recommended
- 29:46portfolios—there are two, appreciation
- 29:48and income—put together and monitored by
- 29:50our team of analysts.
- 29:52You enter how much you have to invest,
- 29:54you see which assets to put it in, how
- 29:55much to put in each, and what the
- 29:57maximum purchase price is.
- 29:58If the asset has passed that price, you
- 30:00know it's not time to buy. And this
- 30:02makes a big difference on days like the
- 30:03ones the Brazilian stock market just
- 30:05lived through.
- 30:06The day after the second round, when the
- 30:08market fell nearly 14%, the Finclass
- 30:09appreciation portfolio fell about 3%.
- 30:12Just as important as knowing where to
- 30:13put your money is knowing where not to
- 30:15put it. You saw in this video the damage
- 30:17that leverage caused in Korea. And
- 30:19inside Finclass, we steer our
- 30:20subscribers away from that path before
- 30:21they even start. In May, Alexandro asked
- 30:23in the community about leveraged ETFs.
- 30:25And the response from Vinicius, one of
- 30:27our moderators, was short: it's best you
- 30:29stay away. And when the market drops,
- 30:30guidance also makes a difference. In
- 30:32July, the same month the Korean market
- 30:34crashed, Diego, a Finclass subscriber,
- 30:36wrote in the community that he was
- 30:38watching his stocks fall sharply and had
- 30:40already thought about selling, but he
- 30:41held on. In his words, it's impossible
- 30:43for so many studious people to see a
- 30:45future in this and be wrong. And on the
- 30:47day he wrote that post, the stocks had
- 30:49already recovered all their losses. And
- 30:50that is the kind of guidance you get
- 30:52when you join Finclass. And man,
- 30:53Finclass goes way beyond just portfolios
- 30:56. You know all that stuff that appeared
- 30:57in this video? US interest rates,
- 30:59inflation, the dollar, elections
- 31:00affecting the stock market. Inside, you
- 31:02have economic courses to help you
- 31:03understand all of this at your own pace
- 31:04.
- 31:06There are over 100 pieces of content,
- 31:07including classes, series, documentaries
- 31:09, and "finbooks," covering everything
- 31:10from your first investment to stocks,
- 31:12real estate funds, fixed income,
- 31:13international investing, crypto, and
- 31:15retirement.
- 31:17And right upon your first access, you
- 31:18receive a study plan that tells you
- 31:20where to start.
- 31:21Even better, when you have a doubt, you
- 31:23have live sessions with our analyst team
- 31:24every business day, plus the community.
- 31:26It's where the moderators answer you,
- 31:28just like you saw in the cases I showed
- 31:30you. Today, there are over 130,000
- 31:31subscribers doing this right along with
- 31:33you. And I have some news for you. On
- 31:35November 27th, Finclass is holding its
- 31:37Black Friday. One day only, with the
- 31:39checkout open for 24 hours. And man, we
- 31:41are preparing a lot for Finclass next
- 31:44year. It’s going to change significantly
- 31:46; it will be much more complete and the
- 31:47price will be adjusted. That’s why this
- 31:49Black Friday will be the last time
- 31:51you’ll be able to subscribe to Finclass
- 31:53for 12 installments of 39.95. Even if we
- 31:56offer a 50%discount in the future, like
- 31:58we usually do on Black Friday, it will
- 31:59be more expensive. Black Friday is on
- 32:02November 27th, one day only, 50%off, 12
- 32:04installments of 39.90, plus a new course
- 32:06on AI for investors and bonuses. The
- 32:08link to join the list is here in the
- 32:10description, or just point your phone
- 32:12camera at the QR code on the screen.
- 32:13Whoever joins the list finds out as soon
- 32:15as it opens, okay? And that's it,
- 32:16cousins. I hope it's become clear what
- 32:18is happening in Korea, why this memory
- 32:20has become the most sought-after piece
- 32:21in artificial intelligence, and why this
- 32:23market will still give us much to talk
- 32:25about. If this video helped you
- 32:26understand this landscape a little
- 32:28better, leave a like and send it to that
- 32:29friend who is always talking about
- 32:31artificial intelligence and the stock
- 32:32market. And tell me here in the comments
- 32:33what you thought, I'll be reading them.
- 32:35Best regards, see you next time, and bye
- 32:36.
About this transcript
This page contains the full transcript of A BOLHA DA COREIA DO SUL ACABOU DE ESTOURAR (explicado) by O Primo Rico, generated from the public captions YouTube serves with the video. The transcript has 6,542 words across 1,024 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.