원문 보기
Hello everyone, congratulations again. Katie Wood told investors in Hong Kong that Ark is avoiding makers of conventional memory chips, saying the trend is up but the cycles are brutal. Micron's latest gross margin was 87%, nearly double last year. So does she see the cycle, or does she miss the contracts?
I'm John A.G. These are Future AI stocks. This video is made for educational and entertainment purposes only. Always do your own research before making any investment decisions . I'm aiming for 500 likes for this video, so hit the like button. Let's get something straight first, because the headlines mix up Katie Wood's two different arguments into one.
Argument number one was made back in August on a podcast. She called high- bandwidth memory (HBM) the most cyclical and most standardized part of the chip supply chain. Her point was that when the price of a part triples, quadruples, or even increases 10 times—this is not a healthy indicator. This is an alarm signal.
And she compared it to Tesla, which removed most of the cobalt from its batteries when it became scarce and expensive. She's betting that chipmakers will do the same with HBM, at least for inference. That's why Ark likes Cerebras and Groq , whose chips rely on fast memory built directly into the chip.
Argument number two was made on October 13 in Hong Kong. Another problem. This one concerns the proposal. She said HBM is currently in short supply, but if more competitors emerge, especially from China, price fluctuations will become greater. State-owned company CXMT is the name that comes up in the conversation, and it has announced plans for server-grade DDR5 by the end of this year.
Now notice what she doesn't say. She is not claiming that the AI boom is over. It's a question of who profits: companies with pricing power or customers who pay the price. This is the right way to test her theory, and that's what we're doing today. Now let's talk about the last two trading sessions, as they gave us a mini test of her argument.
On Thursday, Micron closed at $856.49, down 6.26%, while the Nasdaq fell 2.6%. It is easy to perceive this as the influence of macroeconomic factors, because for the most part it was. Friday is a cleaner test. The NASDAQ fell only 0.2%. The pressure has decreased. Micron rebounded 2.46% to $ 877.55.
But take a look at the storage sector . Western Digital closed at $343.67. A 2.97% drop, even after they beat expectations and raised forecasts on Thursday night. That's about 9.2% lower than Wednesday's close. SanDisk fell 1.82% to $ 1,238.53. So why do stocks fall on good reports when the market calms down?
Two things this week point to the answer. Western Digital's CFO told analysts that input costs for NAND and DRAM are rising, meaning cost savings per terabyte will be lower than the 10 % annual target. The latest TrendForce report suggests that prices are still rising, but suppliers are divided on whether further increases are possible.
Spot trading is sluggish and the growth rate is likely to slow. That's the whole point. This is effectively Katie Wood's argument about cycles, as shown on real-time charts. The market doesn't ask if there is demand. He asks whether the pace of change has reached its peak. When prices are rising so quickly, a slowdown in the rise can feel like a decline.
There is one more nuance. CXMT is a story about DRAM. SanDisk manufactures NAND. The market doesn't always separate these things, and that's something to keep in mind when you see these stocks moving in sync. Here, I believe Wood deserves an unbiased review, and the support comes not only from skeptics.
Some of this comes from within the industry. SK Group Chairman Chey Tae- won, whose company owns SK Hynix, himself called current memory prices abnormally high and said the industry should focus on increasing supply, not squeezing out every last dollar of profit. When the head of a supplier company says that the prices look unusual, it's not just side talk.
Now let's look at the consumer side. SanDisk's consumer revenue fell 32% from the previous quarter as high prices weighed on demand . Think about what this means. People who pay for phones, laptops, and gaming equipment are the first to feel it. When the price increases so quickly, customers do what they always do.
They cut costs, wait, or redesign to avoid this. And this is the engineering part of her argument. Micron says it doesn't have a clear understanding of when supply and demand will balance. Okay, but that same shortage is the strongest incentive for a chip designer to use less memory. Cobalt is an example.
Scarcity not only raises prices, it also teaches consumers to consume less. And now—a nuance. Wood's argument about design exclusions applies specifically to data inference, not to every AI workload. And Micron says HBM shipments will grow faster than conventional DRAM through 2028. So, her thesis may be correct about direction, but the real question is about speed.
Being first and being wrong look the same for a very long time. Now let's turn the situation around. The strongest argument against Wood is the contract book, and the numbers are big. Let's start with the margin it points to. Micron's gross margin ( non-GAAP) last quarter was 87%, up from 45.7% a year earlier .
SanDisk's GAAP gross margin increased from a low of 22.5% to 84.6%. Wood's logic is simple. Margins that rise so quickly thanks to prices can fall just as quickly when supply catches up with demand. Micron's answer is contracts. The company has signed 26 strategic customer agreements covering more than 35% of its projected revenue by 2030, backed by $32 billion in customer commitments, mostly in the form of cash deposits.
These are “take or pay” type deals with minimum price thresholds. The remaining liabilities, calculated at minimum prices, amount to approximately $150 billion. And management says that more than three- quarters of production for fiscal year 2027 has already been contracted. SanDisk has a similar situation.
Its new business model agreements are expected to cover more than half of its production in fiscal year 2027 and about two- thirds in fiscal year 2028, backed by financial guarantees worth $16.5 billion. Here is the part that is important for the discussion. Contracts change the nature of the recession.
They are not canceling it. Three reasons. First, coverage above 35% still leaves a significant portion of income dependent on market prices. Second, RBC noted that no one has tested whether these price thresholds will hold during a real downturn cycle. They were never used under pressure. Third, a take-or-pay agreement is only as strong as the client who signed it.
This week's sell-off was partly due to how AI companies are funding their development. If a client's funding is reduced, the contract may become negotiable. Micron's CFO said that even at rock-bottom prices, margins should still remain well above the peaks of any previous cycles. If this is true, the next recession will look very different from previous ones.
This is exactly the kind of statement the market is supposed to take at face value, and it has never been tested. So how do we keep score? Here's what each side should see over the next few months. Camp Wood will receive support if prices continue to rise, but at a slower pace. If suppliers talk about increasing capacity faster than demand grows.
If CXMT's DDR5 server chips appear in real volumes. And if chipmakers continue to reduce the need for external memory for each AI system, the pro- contract camp will gain support if the number of transactions continues to grow. If suppliers maintain discipline in their spending. And if customers continue to fulfill their volumes under take-or-pay agreements, even when prices stabilize.
And the calendar will provide us with the first data in the coming days. SK Hynix reports on Tuesday evening US time. Listen to how they talk about increasing capacities, because it is this suggestion reaction that triggers every memory cycle. Seagate reports on Tuesday after the close of trading. The first data slice on the storage sector after Western Digital's reaction.
SanDisk reports on Thursday after the close of trading. Revenue is forecast to be between $10.3 billion and $10.8 billion, and earnings per share are expected to be between $44 and $46. The planning has changed . The most important thing may be the forecast for the next quarter and what proportion of the business falls under these multi-year agreements.
The launch of CXMT's DDR5 server memory is a long-term process . This is planned for the end of the year, but translating the design into production output and volumes takes time. This is the real test of the proposition within the arguments about China. Risk one: cycles may last longer than skeptics expect.
Industry analysts do not expect significant supply relief before 2028. A thesis may be correct in direction but wrong in time. Risk two: contracts are protection, not immunity. Minimum price levels have not been verified. Unsecured income depends on the market, and each deal depends on the client's ability to continue paying .
Risk three: design abandonment can happen slowly, then quickly. If more AI chips find ways to use less external memory, the demand structure could change. When developers find a workaround, they often don't go back to the expensive version. Risk four: expectations are ahead of numbers. Friday showed that overshooting the benchmark could lead to a fall if the bar was raised.
Next week's earnings reports could swing these stocks in either direction, despite the long-term arguments. And to be clear, this is not a call to action on what anyone should do with their position. It's about understanding arguments. Here are some frameworks to consider. Wood and the memory manufacturers could be right at the same time, just at different times.
Micron: The contract portfolio is a test of whether this cycle is structurally different. SanDisk: forecast for next quarter. Thursday is a day to watch out for. SK Hynix: supply response speed and how they will talk about capacity this week. Western Digital: Friday showed that success does not always overcome a raised bar.
Seagate: The next data storage. Comparative report on Tuesday. The real debate is between cycle risk and contract protection—and which will show up in the numbers first. This is a complete review of the arguments for and against Katie Wood's memory forecast, built on actual market events this week. If this helped you understand the difference between circular arguments and contract arguments, please click the " like" button.
Leave a comment and tell me whose argument is stronger right now : the cycle or contracts. I read everyone. And if you haven't subscribed yet, now is the time. I am John AG. See you next time.
Comments 0