"뉴스 팔기"인가, 사이클 정점인가? 삼성·SK하이닉스 주가 폭락 이면의 진짜 논리

Compiled & Translated by: Odaily TechFlow

Guest: EJ, Co-host of the Limitless Podcast

Host: Josh, Limitless Podcast

Podcast Source: Limitless Podcast (formerly the Bankless channel)

Original Title: The AI Trade Everyone's Getting Wrong

Air Date: July 9, 2026

Key Takeaways

Samsung's latest quarterly profit reached $58.5 billion, surpassing NVIDIA's $53 billion for the same period, with over 94% of that profit coming from one division: AI memory. Only three companies globally can produce HBM (High Bandwidth Memory): two in South Korea (Samsung and SK Hynix) and one in the US (Micron). Every AI inference requires re-reading the entire model weights, leading to memory demands 10 to 20 times higher than consumer-grade products, and this demand doubles with each new model generation. Meanwhile, producing 1 GB of HBM consumes the equivalent capacity needed for 4 GB of standard DRAM, directly squeezing the memory supply for phones and computers. The price hikes across Apple's entire product line are a direct consequence.

The paradox is this: memory companies are reporting record profits and prices are still rising, yet their stock prices have collectively entered a bear market. Samsung fell 9% on its earnings day, and SK Hynix dropped 15%. The hosts attribute this to the "sell the news" event combined with cyclical panic, stating that the fundamentals remain unchanged. SK Hynix is listing on Nasdaq via an ADR, raising approximately $30 billion with 4x oversubscription, and Leopold Aschenbrenner participated as a seed investor.

Highlighted Insights

Profit margins from the memory monopoly dwarf everything else.

"Samsung's gross margin is 52%, SK Hynix's is 72%. For every $100 of memory sold, $72 goes directly to the bottom line. For comparison, Apple's hardware gross margin is around 30%."

"Samsung memory division employees received year-end bonuses equivalent to 6 times their annual salary. Luxury goods sales in South Korea have tripled over the past 4 months."

AI is a black hole for memory.

"AI is essentially a memory black hole, more extreme than any consumer or enterprise technology wave we've ever seen."

"Every time you submit a prompt, whether it's ChatGPT or Claude, it has to re-read the entire model weights. Every single time."

"Each new model requires 10 to 20 times more memory than its predecessor. With models having 15 trillion, 20 trillion parameters, this demand only goes up."

The price surge isn't over yet.

"Prices rose 90% in Q1, another 50-60% in Q2, and are expected to rise another 20% in Q3. Samsung has made more money in one year than in the last 40 years combined."

Record profits, yet stocks are in a bear market.

"Samsung beat earnings expectations on its earnings day, posting a quarterly profit higher than NVIDIA's. Then its stock dropped 9%. SK Hynix fell 15%."

"Meta hinted at cutting back on AI capital expenditure. Honestly, they haven't really produced anything useful."

"If you bought a month ago, it looks pretty stupid now. But 6 to 24 months down the line, these companies are still very strong businesses."

Main Text

Samsung: The Misunderstood King of AI Profits

Josh: Samsung just became the most profitable company in the world, but almost no one knows exactly how they make their money. I used to think of Samsung as smartphones and computers, but 96% of their profit actually comes from one division: memory. The memory market has only three major suppliers, and coincidentally, two of them are in South Korea. The two most profitable companies in the world are in the same country. What's going on here, EJ? How did Samsung do it?

EJ: When I was a kid, I only knew Samsung as a phone seller. Over the past year and a half, I keep hearing their name tied to AI, and I was curious about what this company actually does. Let's look at some numbers. Samsung's Q2 profit was $58.5 billion, beating the analyst estimate of $55 billion by $3 billion. More importantly, they crushed NVIDIA, which only had $53 billion in the same period. A year ago, in Q2 2025, Samsung's profit was only $3.4 billion. From $3.4 billion to $58.5 billion in one year – what happened?

The answer is HBM, High Bandwidth Memory. Samsung is currently the world's second-largest HBM supplier. Why is HBM so valuable? Because every GPU NVIDIA sells, every TPU Google makes, any AI chip needs a massive amount of memory. Note the word "massive." The amount of memory needed for consumer devices (laptops, computers) is predictable, but the demand for AI memory is exponential.

Samsung makes $650 million a day, $27 million an hour, and $7,500 a second. That's more than the world's most valuable companies. Yet Samsung isn't the world's most valuable company. That's interesting.

The Memory Triopoly and HBM Monopoly

Josh: A computer needs two things: a workbench (memory) and a filing cabinet (storage). Only three companies in the world can make these.

EJ: Right. Three types of memory, each with its own purpose. DRAM is what's in your computer as DDR5 – the workbench, temporary RAM, wiped when you turn it off. NAND is your SSD, the flash storage in your iPhone – the filing cabinet, slower but cheaper. Then there's HBM, High Bandwidth Memory, and this is what the entire AI era is built on.

Manufacturing HBM is incredibly complex. DRAM chips are stacked like skyscrapers, 12 to 16 layers high, using extremely precise processes. SK Hynix holds a 60% share of the entire HBM market. Any AI chip being made is likely using their memory. A triopoly has formed, and profit margins are inflating as a result.

Josh: Here's a very intuitive number. 1 GB of HBM consumes the wafer capacity equivalent to 4 GB of standard DRAM. This means for every wafer shifted to AI memory, the capacity for phone and computer memory shrinks by a factor of four. Apple raised prices across its entire lineup for the first time: MacBook Air from $1100 to $1300, MacBook Pro from $1700 to $2000, Mac Studio from $4000 to $5300. Because people buying a Mac Studio want to run local inference, which has massive memory demands, and there simply isn't enough memory.

AI is a Memory Black Hole

EJ: If I had to explain in one sentence why everyone is so bullish on these companies: AI is essentially a memory black hole, more extreme than any technology wave we have ever seen.

You type a prompt and submit it, whether it's ChatGPT or Claude. Each time, it has to re-read the entire model weights. What are model weights? They are the parameters those companies spent billions of dollars training. Every single inference re-reads them. And each new model requires 10 to 20 times more memory than the last. Models with 15 trillion, 20 trillion parameters – the demand will only keep pushing higher.

But that's not even the biggest part. Your chatbot remembering what you said last time, or retaining your information across conversations – that temporary storage falls under NAND flash memory, and its demand is equally huge. So, demand for all three types of memory is being pulled simultaneously.

Some say this is a bubble that will burst. Historically, there's reason for that; the memory industry has always been cyclical. Three years ago, SK Hynix was nearly acquired by Micron – that was the absolute bottom of the cycle. They had invested in HBM, unsure if they could sell it, and Micron almost bought them. In the end, they didn't sell, doubled down on HBM, and now they are the most valuable company in South Korea.

The Price Surge: A Stepladder from 90% to 20% Increases

EJ: Let's look at the pricing. Over the past 6 months, SK Hynix and Samsung's pricing went up 90% in Q1. The memory inside your everyday phone or computer suddenly doubled in price because one of its components surged nearly 100%. Then another 50 to 60% increase in Q2. Samsung is planning another 20% increase for Q3.

Samsung made more money in one year than it did in the previous 40 years combined, which is 19 times its profit from the same period last year.

In terms of gross margins: a grocery store makes $3 on every $100 of goods sold, an automaker makes $7, Apple hardware makes $30. Samsung's is 52, SK Hynix's is 72. For every $100 of memory sold, $72 is pure gross profit. Samsung memory employees received year-end bonuses equal to 6 times their annual salary. Luxury goods sales in South Korea have tripled in the past 4 months. In Taiwan, people are taking out high-interest loans of $60,000 from banks to buy TSMC stock. Some crazy things are happening in the Asian AI market.

Josh: This price increase is brutal. Consumers are already feeling the pain. A 32GB RAM stick is 2 to 3 times more expensive than it was last year. When building a computer, one-third of the cost is now memory. These prices have already passed through to the actual consumer market. The question is: how long can this last? Can they keep raising prices forever?

EJ: It depends on one variable: whether the number of people using AI continues to grow. This is the only proxy metric to measure whether memory demand will keep increasing. If you believe that in the future, everyone will run multiple AI agents and use AI for work and life, then the demand for memory is exponential.

What about the supply side? New fabs won't be operational until 2030. These factories are incredibly complex to design and build; you can't suddenly flood the supply. Demand is outpacing supply by a factor of 3 to 5, and we are looking at supply constraints for the next few years. In China, CXMT is producing similar DRAM and HBM, but all their capacity is being consumed by Chinese domestic AI labs. Apple tried to find alternative suppliers in China but couldn't get any stock.

There's also a theory: what if a new model architecture emerges that doesn't require as much memory? I think the logic is actually the opposite. As memory gets cheaper, more use cases become viable, more AI agents get deployed, and the overall economic output increases. This leads to an even greater demand for memory.

Paradox: Record Profits, Stocks in a Bear Market

Josh: We've been very bullish on memory. Micron is up 150% since we recommended it late last year. But recently, all memory stocks have fallen more than 20% from their highs. A technical bear market. Samsung beat earnings on its earnings day, posting a quarterly profit higher than NVIDIA's, and then it dropped 9%. SK Hynix fell 15%. Profits are at all-time highs, prices are still rising, but the market is saying "slow down."

What might have spooked the market is Meta hinting at cutting back on AI capital expenditure. Although, honestly, they haven't really produced anything useful. This is the strange point we are at now: companies are saying, "We're doing great, record profits, strong demand," and the market is saying, "Wait, it's too much, there are unknown risks." What do you think?

EJ: My view is simple: this is sell the news. Quarterly earnings season just ended. Global funds had heavy positions in these stocks and were looking for a good price to re-enter. You can think I'm just stubborn if you want. But if you are a long-term AI investor, memory is a necessity, and only these three companies can produce it. This dynamic won't change in the short term.

People are worried about the cycle. During the last memory super-cycle in 2017-2018, Micron's P/E ratio dropped to 4-5x, and then its stock fell 60%, even though profits were still rising. History does seem to be repeating itself, but the difference this time is: last time, it was driven by smartphones, and you could predict the ceiling on demand. This time, it's driven by AI, and you can't see the ceiling on demand.

There's another interesting historical detail. At the bottom of the last cycle, when memory prices were at their cheapest, who was aggressively driving down prices and hoarding memory? Apple. Apple had pricing power over Samsung and SK Hynix, forcing them to supply DRAM and NAND at the lowest prices. Now the tables have turned. Samsung and SK Hynix are doing what Apple did back then. In an efficient market, this is perfectly normal.

SK Hynix Lands on Nasdaq

Josh: Next up is a major test. SK Hynix is a South Korean company, but on July 10th, it is listing on Nasdaq via an ADR, raising approximately $30 billion. This is a crucial moment for the memory industry. How the US market prices SK Hynix will determine the next phase of this memory trade. EJ, will you participate?

EJ: Short answer: Yes. I am bullish on memory. I hold Micron and I also hold a DRAM ETF (a basket of memory companies). If you are a US citizen, it's inconvenient to buy Korean stocks directly, but you can get exposure through these basket products. I've been waiting for SK Hynix to come to the US market for a long time, and I will buy.

Rumors suggest this IPO is 4x oversubscribed. Institutions, pension funds, and retail investors are all trying to get in. Achieving this scale tells you that institutions have done deep research on the memory sector and intend to hold for the long term.

Josh: Guess who contributed $2 to $3 billion of that $30 billion? Leopold Aschenbrenner. He's back again. Last time we did an episode analyzing his positions, some people questioned his logic for shorting NVIDIA. Since that episode, NVIDIA has fallen 20%. Now he's participating as a seed investor in SK Hynix's IPO. You can question his judgment, but he hasn