Compiled & compiled by: Shenchao TechFlow

Guest: EJ, Co-host of the Limitless Podcast

Host: Josh, Limitless Podcast

Podcast source: Limitless Podcast (formerly Bankless channel)

Original title: The AI Trade Everyone's Getting Wrong

Broadcast date: July 9, 2026

Key Takeaways

Samsung's latest quarterly profit is $58.5 billion, surpassing NVIDIA's $53 billion for the same period, with over 94% of profits coming from one division: AI memory. There are only three companies in the world that can produce HBM (high-bandwidth memory), two in South Korea (Samsung and SK Hynix) and one in the United States (Micron). Every inference in AI requires reading the entire model weights, and the memory demand is 10 to 20 times that of consumer-grade products, with each generation of models doubling in demand. Meanwhile, 1 GB of HBM consumes the capacity equivalent to 4 GB of ordinary DRAM, directly squeezing the memory supply for phones and computers, leading to price hikes for all Apple products.

The contradiction lies in the fact that memory companies are posting record profits, prices are still rising, but stock prices have collectively fallen into a bear market. Samsung's stock dropped 9% on earnings day, and SK Hynix fell 15%. Both hosts believe this is a combination of a "sell the news" event and cyclical panic, with no fundamentals changing. SK Hynix listed on NASDAQ in the form of ADR, raising approximately $30 billion, with a 4x oversubscription, and Leopold Aschenbrenner participated as a seed investor.

Highlights of Opinions

Monopoly profits in memory crush everything

"Samsung’s gross margin is 52%, while SK Hynix is at 72%. For every $100 of memory sold, $72 goes directly to the balance sheet. In comparison, Apple's hardware gross margin is about 30%."

"Samsung's memory department employees receive year-end bonuses that are 6 times their annual salary. The luxury goods market in South Korea has seen sales triple over the past four months."

AI is the black hole of memory

"AI is basically a black hole for memory, more exaggerated than any consumer or enterprise technology wave we have seen."

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

"Each new model requires 10 to 20 times the memory of the previous generation. Models with 150 trillion parameters and 200 trillion parameters will only continue to increase this demand."

The price surge is not over yet

"Prices rose 90% in Q1, another 50 to 60% in Q2, and are expected to rise another 20% in Q3. Samsung is earning more in a year than in the past 40 years combined."

Record profits, stock prices in bear market

"Samsung's earnings day beat expectations, with quarterly profits exceeding NVIDIA's. Then the stock fell 9%. SK Hynix fell 15%."

"Meta hinted at tightening AI capital expenditures. To be honest, they haven’t produced anything useful."

"If you bought a month ago, it seems pretty foolish now. But 6 to 24 months from now, these companies will still be strong businesses."

Main Text

Samsung: The Misunderstood King of AI Profits

Josh: Samsung has just become the most profitable company in the world, but almost no one knows how they are making money. I used to think of Samsung only as a smartphone and computer maker, but in reality, 96% of their profits come from one department: memory. There are only three major suppliers in the memory market, two of which happen to be in South Korea. How did the two most profitable companies in the world end up in the same country? EJ, how did Samsung do it?

EJ: When I was young, I only knew that Samsung sold phones. Over the past year and a half, I've constantly heard this name associated with AI, and I am also curious about what this company is doing. Let's look at some numbers. Samsung's Q2 profit is $58.5 billion, with analyst expectations of $55 billion, exceeding expectations by $3 billion. More importantly, they crushed NVIDIA, which only had $53 billion in the same period. A year ago, Samsung's Q2 2025 profit was only $3.4 billion. What happened in that year?

The answer is HBM, high-bandwidth memory. Samsung is currently the second-largest HBM supplier in the world. Why is HBM so valuable? Because every GPU sold by NVIDIA, every TPU made by Google, and any AI chip requires a large amount of memory. Note the word "large," as the amount of memory needed for consumer-grade 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. This is more than the most valuable company in the world makes. But Samsung is not the most valuable company in the world, which is interesting.

The Three Giants of Memory and HBM Monopoly

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

EJ: That's right. There are three types of memory, each with its own use. DRAM is your DDR5 memory inside your computer, a workbench, temporary RAM that is emptied when powered off. NAND is the SSD and flash memory in your iPhone, a filing cabinet, slower but cheaper. Then there's HBM, high-bandwidth memory, which is essential in the AI era.

HBM manufacturing is extremely complex. DRAM chips are stacked 12 to 16 layers high, using extremely precise processes. SK Hynix controls 60% of the entire HBM market. Any AI chip produced is likely using their memory. The monopoly structure of the three companies has formed, leading to inflated profit margins.

Josh: There’s a very intuitive statistic here. 1 GB of HBM consumes wafer capacity equivalent to 4 GB of ordinary DRAM. In other words, for every wafer that shifts to AI memory, there is a 4-fold loss of mobile and computer memory capacity. Apple's first full-price increase has seen the MacBook Air go from $1100 to $1300, the MacBook Pro from $1700 to $2000, and the Mac Studio from $4000 to $5300. Because those buying Mac Studios need to run local inference, the memory demand is huge, and the memory simply isn't enough.

AI is the Black Hole of Memory

EJ: To explain in one sentence why everyone is so optimistic about these companies: AI is fundamentally a black hole for memory, more exaggerated than any technology wave we have seen.

When you write a prompt and submit it, whether ChatGPT or Claude, it has to reread the entire model weights each time. What are model weights? They are parameters that those companies spent billions to train. Every inference requires a reread. Furthermore, every new model's memory requirement is 10 to 20 times that of the previous generation. Models with 150 trillion parameters and 200 trillion parameters will only continue to push this demand upward.

But that's not the biggest part. Your chatbot remembers what you said last time, retaining your information across conversations, and this temporary storage comes from NAND flash memory, which also has huge demand. Thus, the demand for all three types of memory is simultaneously exploding.

Some say this is a bubble that will burst. Historically, there is some truth to this, as the memory industry has always had cycles. Three years ago, SK Hynix was almost acquired by Micron, which was a definitive bottom of the cycle. They invested in HBM without knowing if they could sell it, and Micron almost bought them. In the end, they didn’t sell, they doubled down on HBM, and now they have become Korea's highest-valued company.

Price Surge: From 90% to 20% Rise Steps

EJ: Let's look at the prices. SK Hynix and Samsung’s pricing over the past 6 months: Q1 rose 90%. The memory in your phone and computer suddenly doubled in price, just because one component rose nearly 100%. Q2 rose another 50 to 60%. Samsung expects further increases of 20% in Q3.

Samsung is earning more in a year than in the past 40 years combined, 19 times that of the same period last year.

In terms of gross margin, grocery stores earn $3 for every $100 of goods sold, car manufacturers earn $7, and Apple hardware earns $30. Samsung is at 52, SK Hynix at 72. Every $100 of memory sold results in $72 directly booked. Samsung memory employees receive year-end bonuses that are 6 times their annual salary. The luxury market in South Korea has tripled over the past four months. In Taiwan, someone took out a $60,000 high-interest loan from the bank to buy TSMC. Crazy things are happening in the Asian AI market.

Josh: This increase is too severe. Consumers are already beginning to feel the impact. A 32GB memory stick is 2 to 3 times more expensive than last year. Building a computer, one-third of the cost is just memory. These prices have already propagated to the real consumer market. The question is: how long can this continue? Can they keep raising prices?

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

What about on the supply side? New wafer factories won't be operational until 2030. These factories are designed extremely precisely; they cannot flood supply in the short term. Demand is 3 to 5 times faster than supply, and for the next few years, it will be supply-constrained. In China, CXMT is working on similar DRAM and HBM, but their capacity is all consumed by local AI labs. Apple’s attempts to find alternative suppliers in China have also been fruitless.

Another argument is that if a new model architecture emerges that doesn't require so much memory? I think the logic is exactly the opposite. If memory becomes cheaper, more scenarios can run, more AI agents are deployed, and overall economic output rises. The demand for memory will actually be larger.

The Contradiction: Record Profits, Stock Prices in Bear Market

Josh: We have always been optimistic about memory; Micron has risen 150% since our recommendation last year. But recently, all memory stocks have dropped over 20% from their highs. A technical bear market. Samsung's earnings day beat expectations, with quarterly profits surpassing NVIDIA, then dropped 9%. SK Hynix fell 15%. Profits are at a record high, prices are rising, but the market is saying, "Slow down."

What may have scared the market is Meta hinting at cutting AI capital expenditures. But to be honest, they haven't produced anything useful. This is the strange point in time we are at: companies are saying "We are doing well, profits are at a record high, demand is strong," while the market says "Wait a minute, the increase has been too much, there are unknown risks." What do you think?

EJ: My judgment is very simple: this is just sell the news. The quarterly earnings season just ended, and global funds have heavy positions in these stocks, waiting for a good price to re-enter. If you think I'm just stubborn, I can understand. But if you're a long-term investor in AI, memory is a necessity, and there are only these three companies that can produce it; this situation will not change in the short term.

People are worried about the cycle. The last memory supercycle from 2017 to 2018 saw Micron’s PE drop to 4 to 5 times, then fall 60%, even though profits continued to rise. History is indeed repeating itself, but the difference this time is: last time it was driven by mobile phones, and you could predict the demand ceiling. This time, it's driven by AI, and the ceiling for demand is not visible.

Another interesting historical detail. During the last cycle’s bottom, when memory prices were cheapest, who was frantically squeezing prices and hoarding memory? Apple. At that time, Apple had pricing power over Samsung and SK Hynix, forcing them to supply DRAM and HBM at the lowest prices. Now the tables have turned, and Samsung and SK Hynix are doing what Apple did back then. In an efficient market, this is quite normal.

SK Hynix's NASDAQ Listing

Josh: Next, there's a major test. SK Hynix is a South Korean company but will list on NASDAQ as ADR on July 10, raising about $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 own Micron and also hold a DRAM ETF (a basket of memory companies). If you are a US citizen, directly buying Korean stocks is inconvenient, but you can allocate through these basket products. I have been waiting for SK Hynix to enter the United States, and I will buy.

Currently, it is reported that this IPO was 4 times oversubscribed. Institutions, pension funds, and retail investors are all competing. Being able to reach this scale indicates that institutions have done deep research on the memory sector and plan to hold long-term.

Josh: Guess who will contribute $2 to $3 billion of that $30 billion? Leopold Aschenbrenner. He’s back again. Last time we analyzed his holdings, some questioned his logic in shorting NVIDIA. As it turned out, NVIDIA has fallen 20% since we recorded that episode. Now, he is participating as a seed investor in SK Hynix's IPO. You can question his judgment, but he has indeed never missed.

Long-Term Perspective: Supercycle or Bubble?

Josh: Let’s align on the conclusion. In the long term, memory demand has no ceiling, and only these three companies can produce it. New entrants will not be able to form meaningful competition in the short term. In the short term, people are scared because of the historical cycles casting shadows, and prices have risen too much too quickly. This makes total sense. If you bought a month ago, it looks quite foolish now. But 6 to 24 months from now, the fundamentals of these companies will not change.

EJ: Yes. Ultimately, there will be a cycle top, and that will certainly be when wafer capacity starts becoming surplus. But supply will not ramp up until 2030, while demand is 3 to 5 times supply. Before then, margins and profit margins will continue to expand.

I’m more curious about what the audience thinks. AI memory isn't the sexiest topic, but it is becoming the most important component in the capital expenditure race for AI infrastructure. It is the source that drives up all consumer good prices. Do you think we are too optimistic? Can margins continue to rise? Or is this just a bubble that will eventually burst, while Josh and I are marking the peak?