Effed Orff retweeted
삼전과 닉스의 동상이몽.
같은 반도체 사이클을 타고 있지만, '주주환원'을 대하는 두 회사의 속사정은 완전히 딴판이다.
1. 하이닉스의 꿈: 2029년까지 HBM 장부를 채워둔 상태에서 90조 현금 환원과 자사주 소각을 던졌다. 모회사(SK스퀘어) 지분율이 높아 자사주를 태울수록 대주주 지배력이 강화되니, 주주와 대주주가 같은 꿈을 꾸며 달린다.
2. 삼전의 현실: 84.4조 공장 설비투자에 돈이 묶여 잉여현금(FCF)이 말랐고 환원안을 1월로 미뤘다. 결정적으로 삼성생명·화재 지분율(9.88%)이 금산법 10% 천장에 걸려 있어, 보통주를 소각하면 대주주가 강제로 주식을 매도해야 하는 역설에 갇혀 있다.
3. 결국 승부는 11~12월 수급이다. 1월 말 삼성의 환원 발표와 닉스의 역대급 성과급 잔치가 같은 주에 맞물린다. 그때까지 주가가 올라 있어야 '기다린 보람'이 되지, 흘러내리면 감정 청구서만 남는다. 연말 외국인 매수 연속성과 국민연금의 비중 결정이 핵심 열쇠다.
* 로켓슬라의 네프콘을 요약한 글이며, 미묘한 뉘앙스차이가 있을 수 있는 각색본입니다.
출처: naver.me/G2Y72SDq
@Rocketesla_KR #삼전닉스 #삼성전자 #SK하이닉스
Effed Orff retweeted
Gavin Baker on memory LTAs: “You might blow up your entire business and your franchise by breaking an LTA. And that was never the case before”
Talking about memory makers moving buyers onto long-term agreements (LTAs), he works through the game theory of what happens to a buyer who walks away from one:
“Let’s just think about the game theory of breaking an LTA. So there’s four companies that matter at scale. There’s Amazon with their Trainiums, there’s Google with their TPUs, there’s AMD, and then there’s NVIDIA who’s much bigger than everybody else combined.”
“Let’s just say it’s 2027, it’s very important to realize memory is, the more memory you put with FLOP for a given unit of compute, the more tokens you get out. It’s the single most important thing you could do to increase kind of token output per unit of compute. And then that obviously definitionally actually lowers costs, which is why the demand hasn’t responded at all negatively. There’s been no elasticity just because it’s kind of the only, it’s the axis that is dominating all others. And this is at some level, a giant Game of Thrones or emperors between these companies.”
“And okay, it’s 2027 or '28, you’re vaguely tempted to break one of these LTAs, try and get a lower price. But to a large degree, market shares are, I think for the next several years, are going to be determined by supply chain allocations and what you have pre-purchased. So if you break the LTA and you, and this is assuming we’re not in a severe oversupply situation, the game theory even holds in a severe oversupply situation. If you break your LTA, and then in the next two or three years, for any reason, leverage shifts back to the memory guys, you’re out of business. It’s over.”
“Let’s just say Google breaks an LTA. There’s an oversupply. I’m making this up, in '28, '29. They break their LTAs. Well, if they’re breaking their LTAs, it probably means, your oversupply, prices are coming down, and then, capacity naturally contracts. Well, what do you think is going to happen to Google’s allocations? And then, this is a cyclical industry, and oversupply is followed by undersupply. What do you think they think is going to happen to their allocations next time?”
“So I just think given that this is the axis around which kind of everything is revolving, man, you might blow up your entire business and your franchise by breaking an LTA. And that was never the case before.”
I'd like to believe LTAs are durable too. But take one example.
Analog chipmaker onsemi said this in 2023.
On January 5, 2023, at JPMorgan's CES forum, CEO Hassane El-Khoury was asked how much LTSAs would still hold once the shortage eased. His answer: “They are a legally binding agreement ’cause they’re a liability for the customer because they’re take or pay.” He also said he would not accept a customer reopening price talks just because it could get the same parts 10% cheaper elsewhere.
On October 30, 2023, on the Q3 earnings call, onsemi cut its SiC shipment forecast for the year from $1 billion to more than $800 million after one automotive OEM reduced demand. An analyst asked how this could happen when the company had been saying its LTSAs were legally binding. The same CEO reaffirmed that they were binding, but said he did not want to enforce the contracts just to push inventory onto a customer whose demand had dropped. If it worked for both sides, terms could be adjusted customer by customer, whether by a quarter, a year, or longer.
In its 10-K filings, onsemi also states that it amended LTSAs with some strategic customers by mutual agreement in 2024, and changed delivery schedules and committed volumes on some LTSAs in 2025 as demand shifted.
Microchip is another case. Customers in its Preferred Supply Program (PSP) originally had to place 12 months of orders and could not cancel or reschedule them unless Microchip raised prices. On its November 2025 earnings call, however, the CEO said the company had accepted some cancellations and pushouts so customers could adjust their backlog. On the same call, the CFO separately addressed the multiyear LTSAs the company had signed. Some of those contracts were still in effect, he said, but customers had been allowed to defer taking delivery of committed volumes by one to two years if they needed to.
Every time I see cases like these, I doubt how much memory makers' LTAs will protect earnings in a downcycle. Right now the most reliable thing is the prepayments and deposits themselves. That cash can go into capacity expansion today.
I can picture hyperscalers with no DRAM inventory continuing to buy at current prices. But once inventory has piled up, will they keep taking every committed unit at the same price? I have a hard time picturing memory makers holding customers to the high contract price all the way through.
The LTA itself may never be broken and may stay in place. But if both sides agree to push this year's volumes into next year or two years out, this year's earnings estimates come down. If price gets renegotiated too, margin estimates have to come down as well.
Again, I agree LTAs are sturdier than they used to be. What I question is whether, after demand turns, memory makers can still earn 80% margins on contracts signed during the upturn and keep collecting hundreds of billions of dollars in profit because of those LTAs.
Effed Orff retweeted
According to Korean media, Samsung and SK Hynix are now racing each other for ASML's EUV machines.
Each company is reportedly putting between ₩10 trillion and ₩20 trillion into EUV orders.
$ASML has formally told customers it cannot make enough EUV tools for 2027 and 2028.
SK Hynix already disclosed an order of about ₩11.95 trillion, with delivery due by the end of 2027, and both companies are considering ordering EARLIER because lead times have doubled.
Samsung is ordering tools for its new Pyeongtaek P5 fab, which reportedly needs more than ₩60 trillion per fab.
$NVDA just asked its foundry to start making MEMORY.
According to The Information, $TSM has begun drafting plans to produce 3D DRAM after a request from Nvidia, its biggest customer.
TSMC is looking at building new fabs or buying an existing plant and converting it, the same playbook $MU used when it bought a fab and flipped it to DRAM.
The effort is being led by Hsu Kuo-chin, a TSMC VP who oversees advanced packaging and previously worked at Micron.
If this happens, the company that already makes Nvidia GPUs and packages them would also make the memory next to those GPUs.
That could be a direct long term threat to Samsung, SK Hynix, and Micron, who have owned this market for decades.
The memory shortage is so bad that Nvidia is trying to invent a fourth supplier.
Effed Orff retweeted
It think it's stupid that "AI Bubble" gets used over and over. In infra names you have:
- 1-3x fwd P/E memory companies from Samsung to $ESMT. With HBM players now having hyperscaler multi-year price floor LTAs.
- Optical companies like $POET traded -$97M under net cash. Valuing the operating business very negatively, not even at $0.
- While AI companies with larger premiums continue to record demand over the next 3-5 years. Such as the new $20B LTAs signed w/ $COHR to $5B LTAs signed w/ $LITE.
Can someone in the media honestly see negative EV high-beta optical companies, where 2 years of burnrate would still value the company below $0.
or single digit memory companies with contracted demand + price floors years out and say that's a bubble?
Many of these names are extremely profitable today...
But obviously there's other elements inside the theme, such as:
- Certain startups like Instinct that look like a private valuation bubble.
- Leveraged Neoclouds that didn't finance properly
- and AI lab valuations and commitments, which should be viewed cautiously.
But everything gets grouped into one with the boogyman of "leverage" or "debt".
Yes, valuations can compress or correct from macro, potential regulatory uncertainty, or potential capex reductions.
But for the companies physically powering the AI buildout, it looks more like the Great AI Deleveraging of 2026...
Than anything like fundamental demand destruction.
As more earnings come out and free cash flow accumulates over time.
I'm confident in a recovery, especially in bottlenecks, and for markets to eventually discern durability + growth over frothy valuations tied to AI.
Effed Orff retweeted
삼전 전략쪽 친구 말로는,
중국의 생산량 케파 업이나, 하이닉스가 미국/일본/베트남 확장 계획하는걸 삼성도 다 알고 있다고 하더라.
반면에, 삼성이 대규모로 평택을 확장하는 것에 대해서도 하이닉스도 전부 알고 있다고.
이런 저런 케파업 흐름을 고려해도, 29년도까지는 성장을 지속 할 것으로 현재는 전망하고 있어서, 본인은 삼성전자 주가가 내려오면 내려올수록 비중을 더 늘릴거라고 하더라.
현재 시점에서, 필요한 수요대비 모든 메모리반도체 업체들의 공급량 숏티지는 2.25배 정도로 보고 있다고 한다.
이 수요가 현재 속도로 유지될 경우에, 생산량이 2배로 늘어나는 시점이 28~29로 보고 있는데, 그 시기부터 공급 과잉이 발생하면 반도체 시세는 빠르게 하락 할 수 있고 이런 사이클이 3번 정도 있었다고 한다.
그럼에도 불구하고, 만약 현재보다 수요가 증가하는 속도가 더 가팔라진다면 이야기가 달라질텐데, 피지컬 AI쪽에 수요가 증가하는 것은 한 5년 뒤쯤에나 될 것 같다고 보고 있다고 하고..
이제는 AI 때문에 수율 잡는 부분에서 돌파구를 찾거나, 재료 부분에서 돌파구를 찾는 등.. 모든 것들이 너무 빠르게 바뀌고 있어서 어떤식으로 흘러갈지 정확히 예상이 안된다는 의견이었다.
피지컬 AI 쪽에서 실제로 수요가 늘어날 가능성이 있어도, 과연 언제부터 어느정도 규모로 발생 될 것인지도 불투명한 상황인데, 방향은 확실히 터질 것 같은데 그게 메모리 가격에서 실제로 임팩을 끼칠 정도는 언제인지 정확히 알 수 없다는 의견이었다.
레인보우로봇틱스는 기술 능력이 너무 낮다고 생각하고 있고, 한국에서는 로보티즈가 최고인데 회사에서 애초에 잘못판단했었다는 의견도 줬었고, 회생시켜보려고 해도 레인보우로보틱스에서 희망이 안보인다는 의견.
재용사마가 어떤 식으로든 로봇을 활용해보라고 프레셔를 강하게 주신다는데, 잘 안되고 있어서 답답해하는 것 같다고 하고, AI 를 직원들이 더 많이 더 편하게 활용 할 수 있도록 개선해달라는 주문들이 나온다고 했다.
똑똑한 친구에게 이런 저런 세상 돌아가는 이야기를 들으니 감회가 참 새롭기도 했고, 친구가 행복하게 아이와 지내는 이야기를 들으니 기쁘기도 했다.
이 친구가 삼성전자 7만원일 때, 삼성전자는 미래가 없다며 형 당장 파세요라고 권했던 그 친구인데.. ㅎㅎ
여전히 나에겐 많은 추억과 스키 해외 원정을 함께한 동료이다.
그와 그의 새로운 가정이 늘 건강하고 행복하기를.
Effed Orff retweeted
Cyclical money doesn’t count. It’s fake money. Let’s do a fun comparison between Micron and Wal Mart. But I’m going to give Wal Mart a really big advantage. Let’s see what happens.
Let’s look at current cash and 2027 profits. Wal Mart is actually negative $45 billion dollars because they borrow a lot of money to pay for inventory and stuff. But for this exercise let’s count their inventory as cash. Including inventory and all debt, Wal Mart has about $11.5 billion dollars in cash. They will make about $22 billion dollars in profit in 2027. They grow their profits around 5% a year every year. Their market cap is right around $1 Trillion dollars.
Micron has about $68.3 billion dollars in net cash. They will make at least $200 billion dollars in profits in 2027. Their market cap is about $780 billion dollars ! I can’t help but laugh when I type that. Crazy.
Here is the fun part. The current 30 year bond pays 5.26%. Let’s compare Wal Mart growing their cash pile and their profits at 5% a year. By Jan 1st 2040, that cash and total cumulative profits will be worth $401 billion dollars. That’s assuming Wal Mart is still in business and growing at their average growth of 5% a year. That’s cumulatively. All together from now until 2040. That’s assuming Ai doesn’t disrupt them and the world still shops at Wal Mart in 2040. I’m not saying Wal Mart won’t be here. But I am saying anything can happen by then! Just like memory might still be under supplied by then!
Now, let’s get to Micron. Micron has $68 billion in cash and they will make $200 billion in 2027. Let’s pretend Micron closes shop in 2028. They make zero dollars of new money from 2028 to 2040, but they put that original $268 billion dollars in a govt 30 year bond and get 5%. I won’t even bother using the extra .26% let’s just use a flat 5% rate. That $268 billion dollars growing at 5% until Jan 1, 2040 will be worth $481 billion dollars. Thats over $80 billion dollars more than Wal Mart by 2040. That’s nearly 4 full years of Wal Mart’s profits today.
So, Micron could literally shut down their business from 2028 to 2040 and they will still have more cash than WalMart at that same time. I don’t give a fuck about cyclical bullshit. The cash is real and no one has a clue when the so called “cycle” will end. At some point their actual business and cash has to matter.
The idea that these crazy “real” profits that MU is making right in front of our eyes doesn’t matter, doesn’t make any sense. I could understand applying all of the bullshit hypothetical reasoning that they come up with if MU were trading at 50x forward PE, but it’s not. It has way more than enough risk built into the price. Way too much built in, IMO! There is no reason MU shouldn’t trade at an average market multiple of 20x forward PE. Or $3,600 a share.
Why are we crushing a stock doing everything right by pricing in a fabricated future without any evidence of that? Why don’t we price in the actual reality and then punish the stock if it ever deserves to get punished? We give Wal mart the benefit of the doubt for its future, when no company’s future is certain. If Wal Mart has a sudden drop in business and profits sink, the stock will get punished, as it should after that news. But not before it happens! That’s how it has always worked. Why is this time different? Why do we now have psychics that think they can see the future?
I keep saying that I’ve never seen anything like this. I’ve never seen a stock get punished for a completely made up story about its future without one piece of evidence that points to it. Usually it’s the other way around. Usually the market prices in hopes and dreams in the future. Not demise! Make it make sense!
Effed Orff retweeted
HBM is no longer a drop-in commodity. From HBM4 on, the base logic die is co-designed with the customer’s accelerator.
It carries custom interfaces, controllers, and logic blocks built on leading-edge nodes.
$NVDA
NVIDIA’s NVHBM, done with multiple memory suppliers, is one visible case. It targets higher bandwidth and lower power while freeing silicon area on the compute die. Similar custom work is underway with hyperscalers and design partners for next-generation platforms.
Once a customer’s architecture, package, thermal design, and software stack are built around a specific HBM configuration, cutting committed volume is not a simple commercial call. It requires redesign, re-qualification, and schedule slips that can delay an entire system launch by a year or more.
Qualification is platform-specific and slow. Historical timelines from sampling to volume on major AI accelerators have routinely run 9-18 months even when successful. Failures have cost entire product cycles. Switching suppliers mid-generation carries real technical and schedule risk, not just pricing risk.
Memory makers have answered by locking multi-year deals.
$SKHY
SK hynix has signed LTAs with 14 major customers covering more than half of its 2027-2028 DRAM and HBM capacity, with production effectively committed through 2029 when advanced talks are included.
#Samsung has secured agreements covering a large share of future output and reports next-year HBM volumes as sold out. Many of these contracts include deposits or take-or-pay terms.
The demand driver is still hyperscaler AI infrastructure spending. Combined capital expenditure guidance from the largest cloud providers for 2026 already exceeds $700 billion.
Memory is a rising share of server costs, and HBM remains supply-constrained relative to requests.
HBM also consumes roughly three times the wafer capacity of conventional DRAM per bit, so capacity cannot be quickly redirected or inventoried like standard DDR.
In this setup, the practical durability of supply agreements comes less from pure legal wording and more from joint product development, qualification barriers, and continued structural demand for AI compute.
I respect Jukan, but I feel a little frustrated whenever I see this kind of argument.
You can build any scenario around plausible assumptions, and yes, of course, the outcome will follow from those assumptions. If you assume demand collapses, customers accumulate inventory, and contract terms get renegotiated, then naturally, memory companies’ revenue and profits will decline.
But honestly, this feels like the perspective of analysts who treat memory as just a commodity you manufacture and sell, without properly understanding the technological changes taking place in custom memory.
Who exactly are these customers we are talking about when we say they could cancel LTAs or defer deliveries? They are the very customers working with memory companies to design the next generation of HBM and 3D stacked DRAM.
After jointly designing and validating these products, can customers really cancel those volumes so easily or push deliveries out by one or two years? How would they manufacture their own products, which were designed around that memory? What happens to the launch and system deployment schedules they have already coordinated? Switching suppliers also raises questions of design compatibility and requalification. If you are going to argue that contracts can be adjusted, you also need to explain what that decision would mean for the very customers canceling them.
This is one of the biggest reasons I keep saying that memory is no longer the same cyclical industry it used to be. The parties to these LTAs are already tied together from the design stage. But if you still view memory simply as a commodity where rising prices lead to more production, and inventory buildups lead to price cuts or canceled orders, you will inevitably miss this shift in the technical and business relationships.
One more thing. These arguments about a collapse in memory demand seem to assume that planned AI data center investments will be canceled on a massive scale, or that AI infrastructure expansion will slow so substantially that memory demand falls sharply. What interests me more than whether contracts can be adjusted is why you expect demand to weaken that dramatically in the first place.
I think it would be more realistic to discuss directly whether the AI era we expect will actually fail to materialize, and whether much of the AI infrastructure investment currently planned will disappear. If that is your outlook, I can at least understand why you would expect memory demand to collapse and customers to cut volumes even for products they have jointly developed.
Effed Orff retweeted
Totally wrong. Germany has a similar take-or-pay gas contract with Russian and even they cannot get out of it. Memory LTAs in the past were totally different, with no volume requirements, no deposits. This is just more fear mongering during a market correction.
Effed Orff retweeted
The idea that AI capex will eventually peak is so counterintuitive to me
There may be a local top, possibly even a bad year from time to time, but on a structural basis? This industry will do nothing but keep growing
Effed Orff retweeted
As Charlie Munger always said, there are three downfalls to great men
Liquor, ladies and leverage
BREAKING: US intelligence says Nvidia CEO Jensen Huang had a "romantic relationship" with a Chinese spy
20% chance he's out as $NVDA CEO before 2029
Readers added context they thought people might want to know
US intelligence has issued no public confirmation; the allegation is based on media reports citing anonymous sources about a classified memo. Nvidia has described interactions as routine business.
afr.com/technology/us-…
nytimes.com/2026/11/06/us/…
engadget.com/2310084/china-…
Effed Orff retweeted
Quite well written.
There are popular accounts BSting about LTA vulnerability mentioning incidents before 2024. Those are fixed price contracts which don't protect volume. Most take-or-pay contracts in memory business are quite recent. You probably heard Germans were still paying Russians early this year due to take-or-pay LNG contracts. Even with the extreme hostility Germans couldn't walk away.
There are possibilities of renegotiation, although I doubt it will happen, taking DRAM instead of HBM, which won't hurt memory big 3. When there's a infrastructure supply driven downturn, companies would most likely channel their cash into revenue generating services rather than frontier research. Hyperscalers can spend more on DRAMs and SSDs for services than state of art GPUs. They can generate revenue while DC bottleneck sustains.
I wrote this to counter the potential for inconsequential cancellation of HBM orders. Obvs I used AI to structure it, but the arguments are mine and I’m lazy.
Long-term HBM take-or-pay contracts are fundamentally different from traditional commodity DRAM purchasing arrangements. Several factors make default commercially and strategically unattractive.
1. Only three manufacturers: SK Hynix, Samsung and Micron are the only companies producing HBM in significant commercial quantities. A customer cannot easily find an alternative supplier.
2. Loss of future allocation: If XYZCo defaults on its contract, its supplier may reallocate production capacity to other customers. When XYZCo subsequently needs HBM, it risks finding itself at the bottom of a long waiting list, behind customers that honoured their commitments.
3. Strengthening competitors: Capacity surrendered by XYZCo could be allocated to Nvidia, AMD, Google, Amazon or another competitor. XYZCo would potentially be financing its competitors’ growth by surrendering access to a scarce strategic resource.
4. Contractual penalties: Under a genuine, enforceable take-or-pay agreement, XYZCo may remain liable for contracted purchases even if it no longer requires the memory. Default could therefore create substantial financial liabilities without eliminating the underlying cost.
5. Increasing product customisation: From HBM4 onwards, customer-specific logic base dies, packaging and integration make memory progressively less interchangeable. Abandoning a supply agreement could mean losing years of joint development, qualification and engineering investment.
6. An expanding technological moat: Each HBM generation requires increasingly sophisticated fabrication, stacking, packaging, thermal management and integration technologies. China’s inability to purchase EUV lithography equipment, together with other technological restrictions, makes the emergence of additional leading-edge suppliers particularly difficult.
7. Increasingly proprietary architectures: Future technologies such as zHBM and 3D DRAM could integrate memory much more closely with compute, incorporating customer-specific intellectual property. This would increase switching costs and deepen long-term supplier relationships, although these technologies remain under development.
8. Expanding underlying demand: AI training, inference, agentic AI, hyperscalers, neoclouds and sovereign AI infrastructure all compete for computing capacity. Customers may therefore prefer to retain HBM allocations through temporary demand slowdowns rather than risk being unable to expand subsequently.
9. No established HBM commodity cycle: Unlike conventional DRAM, AI-driven HBM demand has not yet experienced a complete boom-and-bust cycle. There is consequently no demonstrated historical basis for assuming that HBM prices and profitability must follow traditional DRAM cycles.
Conclusion
For XYZCo, the cost of defaulting could greatly exceed the cost of honouring its contract.
Default could expose XYZCo to contractual liabilities, destroy its credibility, jeopardise future access to one of the world’s most strategically important semiconductor technologies and hand scarce production capacity to competitors.
The combination of concentrated supply, increasing customisation, formidable barriers to entry and expanding AI demand gives customers powerful incentives to maintain their contractual commitments.
The greatest deterrent may not be the financial penalty for default. It is the prospect of being unable to obtain HBM when the customer needs it again.
This argument is strongest while HBM capacity remains scarce. It does not make defaults impossible, particularly if a customer becomes insolvent or industry-wide oversupply develops. But it provides a compelling reason why long-term HBM contracts could prove considerably more resilient than conventional memory purchasing agreements.
Effed Orff retweeted
Was at a birthday party today chatting to a dad working the legal side of a bunch of these data center deals
His perspective was that the politicians are mostly barking to get elected. Big tech will have to pay their pound (or two) of flesh to communities but the show goes on
Had obviously been working around the clock to lock down what they could over the last few months, but basically felt this was a hiccup and money gets shovels moving
Who knows. Was just some bloke in a quarter zip eating cold pizza his kid didn’t want. Don't do anything stupid.
Effed Orff retweeted
This is how an amateur/inexperienced trader thinks. They look for reasons, but never attribute them to macro factors.
Why would DRAM makers take on LTAs if they didnt have any downside protection. Thats the entire point of them?!
They were created to break the prior cycle.
Minimum purchases, prices floors, back end agreements & as he states.. up front prepayments (which are very useful now)
Incredible to become this bearish with tech trading at 17x forward P/E & nearly all indicators at decade lows.
Lets not forget that Korea has been getting deleveraged consistently for months. I think companies have grown too fast for natural capital to support their fair values. In time it should come. $MU $SKHY
I know Jukan is young but he will be cancelled soon if he doesnt stick to raw research.
I'd like to believe LTAs are durable too. But take one example.
Analog chipmaker onsemi said this in 2023.
On January 5, 2023, at JPMorgan's CES forum, CEO Hassane El-Khoury was asked how much LTSAs would still hold once the shortage eased. His answer: “They are a legally binding agreement ’cause they’re a liability for the customer because they’re take or pay.” He also said he would not accept a customer reopening price talks just because it could get the same parts 10% cheaper elsewhere.
On October 30, 2023, on the Q3 earnings call, onsemi cut its SiC shipment forecast for the year from $1 billion to more than $800 million after one automotive OEM reduced demand. An analyst asked how this could happen when the company had been saying its LTSAs were legally binding. The same CEO reaffirmed that they were binding, but said he did not want to enforce the contracts just to push inventory onto a customer whose demand had dropped. If it worked for both sides, terms could be adjusted customer by customer, whether by a quarter, a year, or longer.
In its 10-K filings, onsemi also states that it amended LTSAs with some strategic customers by mutual agreement in 2024, and changed delivery schedules and committed volumes on some LTSAs in 2025 as demand shifted.
Microchip is another case. Customers in its Preferred Supply Program (PSP) originally had to place 12 months of orders and could not cancel or reschedule them unless Microchip raised prices. On its November 2025 earnings call, however, the CEO said the company had accepted some cancellations and pushouts so customers could adjust their backlog. On the same call, the CFO separately addressed the multiyear LTSAs the company had signed. Some of those contracts were still in effect, he said, but customers had been allowed to defer taking delivery of committed volumes by one to two years if they needed to.
Every time I see cases like these, I doubt how much memory makers' LTAs will protect earnings in a downcycle. Right now the most reliable thing is the prepayments and deposits themselves. That cash can go into capacity expansion today.
I can picture hyperscalers with no DRAM inventory continuing to buy at current prices. But once inventory has piled up, will they keep taking every committed unit at the same price? I have a hard time picturing memory makers holding customers to the high contract price all the way through.
The LTA itself may never be broken and may stay in place. But if both sides agree to push this year's volumes into next year or two years out, this year's earnings estimates come down. If price gets renegotiated too, margin estimates have to come down as well.
Again, I agree LTAs are sturdier than they used to be. What I question is whether, after demand turns, memory makers can still earn 80% margins on contracts signed during the upturn and keep collecting hundreds of billions of dollars in profit because of those LTAs.
Effed Orff retweeted
I respect Jukan, but I feel a little frustrated whenever I see this kind of argument.
You can build any scenario around plausible assumptions, and yes, of course, the outcome will follow from those assumptions. If you assume demand collapses, customers accumulate inventory, and contract terms get renegotiated, then naturally, memory companies’ revenue and profits will decline.
But honestly, this feels like the perspective of analysts who treat memory as just a commodity you manufacture and sell, without properly understanding the technological changes taking place in custom memory.
Who exactly are these customers we are talking about when we say they could cancel LTAs or defer deliveries? They are the very customers working with memory companies to design the next generation of HBM and 3D stacked DRAM.
After jointly designing and validating these products, can customers really cancel those volumes so easily or push deliveries out by one or two years? How would they manufacture their own products, which were designed around that memory? What happens to the launch and system deployment schedules they have already coordinated? Switching suppliers also raises questions of design compatibility and requalification. If you are going to argue that contracts can be adjusted, you also need to explain what that decision would mean for the very customers canceling them.
This is one of the biggest reasons I keep saying that memory is no longer the same cyclical industry it used to be. The parties to these LTAs are already tied together from the design stage. But if you still view memory simply as a commodity where rising prices lead to more production, and inventory buildups lead to price cuts or canceled orders, you will inevitably miss this shift in the technical and business relationships.
One more thing. These arguments about a collapse in memory demand seem to assume that planned AI data center investments will be canceled on a massive scale, or that AI infrastructure expansion will slow so substantially that memory demand falls sharply. What interests me more than whether contracts can be adjusted is why you expect demand to weaken that dramatically in the first place.
I think it would be more realistic to discuss directly whether the AI era we expect will actually fail to materialize, and whether much of the AI infrastructure investment currently planned will disappear. If that is your outlook, I can at least understand why you would expect memory demand to collapse and customers to cut volumes even for products they have jointly developed.
I'd like to believe LTAs are durable too. But take one example.
Analog chipmaker onsemi said this in 2023.
On January 5, 2023, at JPMorgan's CES forum, CEO Hassane El-Khoury was asked how much LTSAs would still hold once the shortage eased. His answer: “They are a legally binding agreement ’cause they’re a liability for the customer because they’re take or pay.” He also said he would not accept a customer reopening price talks just because it could get the same parts 10% cheaper elsewhere.
On October 30, 2023, on the Q3 earnings call, onsemi cut its SiC shipment forecast for the year from $1 billion to more than $800 million after one automotive OEM reduced demand. An analyst asked how this could happen when the company had been saying its LTSAs were legally binding. The same CEO reaffirmed that they were binding, but said he did not want to enforce the contracts just to push inventory onto a customer whose demand had dropped. If it worked for both sides, terms could be adjusted customer by customer, whether by a quarter, a year, or longer.
In its 10-K filings, onsemi also states that it amended LTSAs with some strategic customers by mutual agreement in 2024, and changed delivery schedules and committed volumes on some LTSAs in 2025 as demand shifted.
Microchip is another case. Customers in its Preferred Supply Program (PSP) originally had to place 12 months of orders and could not cancel or reschedule them unless Microchip raised prices. On its November 2025 earnings call, however, the CEO said the company had accepted some cancellations and pushouts so customers could adjust their backlog. On the same call, the CFO separately addressed the multiyear LTSAs the company had signed. Some of those contracts were still in effect, he said, but customers had been allowed to defer taking delivery of committed volumes by one to two years if they needed to.
Every time I see cases like these, I doubt how much memory makers' LTAs will protect earnings in a downcycle. Right now the most reliable thing is the prepayments and deposits themselves. That cash can go into capacity expansion today.
I can picture hyperscalers with no DRAM inventory continuing to buy at current prices. But once inventory has piled up, will they keep taking every committed unit at the same price? I have a hard time picturing memory makers holding customers to the high contract price all the way through.
The LTA itself may never be broken and may stay in place. But if both sides agree to push this year's volumes into next year or two years out, this year's earnings estimates come down. If price gets renegotiated too, margin estimates have to come down as well.
Again, I agree LTAs are sturdier than they used to be. What I question is whether, after demand turns, memory makers can still earn 80% margins on contracts signed during the upturn and keep collecting hundreds of billions of dollars in profit because of those LTAs.
Effed Orff retweeted
TSMC has pushed back on a report that it is exploring 3D DRAM memory production at Nvidia's request, saying it is a pure-play logic foundry and plans to stay one, according to Tom's Hardware.
Effed Orff retweeted
Data centers in Loudoun County, Virginia: “We get $1.2 billion a year in local tax revenue, about 40% of all of our tax revenue...on 3% of our land”
Buddy Rizer, the county's economic development chief, explained how “data center alley” grew:
“We would have never looked at having 55 million square feet of data center, 230-some buildings. That was not the plan. The plan was to utilize the fiber loop, the internet exchange, to utilize a reclaimed water line, and buildings that were already built. That was the plan, right? And so that made a lot of sense.”
“But, you know, fast forward, today we are largely dependent on that. We get $1.2 billion a year in local tax revenue, about 40% of all of our tax revenue. And while that looks overwhelming, data centers take only about 3% of all of our land in the county. So it’s quite a return on investment. You know, 40%, $1.2 billion, on 3% of our land.”
SpaceX’s Arizona data center, per a county supervisor: “We could be reducing property taxes across the board in Cochise County, the entire county, by 20%”
Frank Antenori, whose district includes the site, brought up the county’s tax math during a live Q&A:
“Here’s just something to just chew on. Arizona has an expenditure limit requirement that is placed on counties. If we exceed that expenditure, we have to rebate back to the taxpayers a certain amount.”
“So, if we’re receiving revenue in from SpaceX, by law, we have to offset that to the taxpayers. We have the potential to reduce, based on the projected revenue they’re creating, as much as 20%.”
“Let me just say it again. We could be reducing property taxes across the board in Cochise County, the entire county, by 20% to offset for the revenue here, because we can’t spend all that money. So, we have to give that money back to the taxpayers. The law says we have to.”
“So, you could see a potential significant impact to local property taxes. What that is, I don’t know. We’re going to have to do the math on it. But the law, if we have this large increase in property tax revenue, we have to balance those rates. So that will reduce dramatically the property tax burden of every resident in the county.”
“The construction sales tax, which also is the TPT, the transaction privilege tax, the retail tax, and the jail excise tax. So that will be part of the construction tax. We could literally use that money to pay the jail off in 2 to 3 years, as opposed to the 13 to 20 years.”
“So this could be a revenue to eliminate the half-cent sales tax in 2 to 3 years. We can get rid of that half-cent sales tax.”
Effed Orff retweeted
$MU is below 4 forward PE despite being critical to the AI revolution and robotics later on. SK hynix $000660.KS is even below 3 forward PE which is insane.
And somehow you have stocks like Walmart $WMT at 45 PE and Costco $COST at 50 PE.
Think about how wild the cyclical memory thesis for Micron is. The thesis is that we should ignore the hundreds of billions of dollars in profits that Micron will make as far out as orders can go. They don’t matter. Even if they make $500 to $600 billion dollars of profits over the next 2-3 years. It doesn’t matter. All because one day memory prices will fall so much that they will bring down the margins. No one knows if this will actually happen, but the ones that claim it will, are certain. But, they don’t know what the future margins will be or how much the prices will fall. Yet, they assign basically a complete collapse. All based on past cycles that had nothing to do with today’s business. There was no Ai in the past. The worst part is There literally is no way to prove or disprove this hypothesis. Or when that may happen. Micron can beat earnings for the next 8 quarters in a row and they can keep saying the same thing. Every prediction is a complete guess. These are just baseless predictions without any evidence on today’s current data. If Micron were based on current evidence with all of the information we actually have today, Micron would trade at a premium multiple to the market. Not even an average multiple. Definitely not a discount. It would probably command a 50 x forward PE. If anyone out there can prove me wrong with actual current data, please do. Not past bull shit stories. Not predictions without facts. Real numbers for facts. And not technical charts. Pure real fundamentals. Actual data. I’d love to read what you can come up with.
Effed Orff retweeted
Think about how wild the cyclical memory thesis for Micron is. The thesis is that we should ignore the hundreds of billions of dollars in profits that Micron will make as far out as orders can go. They don’t matter. Even if they make $500 to $600 billion dollars of profits over the next 2-3 years. It doesn’t matter. All because one day memory prices will fall so much that they will bring down the margins. No one knows if this will actually happen, but the ones that claim it will, are certain. But, they don’t know what the future margins will be or how much the prices will fall. Yet, they assign basically a complete collapse. All based on past cycles that had nothing to do with today’s business. There was no Ai in the past. The worst part is There literally is no way to prove or disprove this hypothesis. Or when that may happen. Micron can beat earnings for the next 8 quarters in a row and they can keep saying the same thing. Every prediction is a complete guess. These are just baseless predictions without any evidence on today’s current data. If Micron were based on current evidence with all of the information we actually have today, Micron would trade at a premium multiple to the market. Not even an average multiple. Definitely not a discount. It would probably command a 50 x forward PE. If anyone out there can prove me wrong with actual current data, please do. Not past bull shit stories. Not predictions without facts. Real numbers for facts. And not technical charts. Pure real fundamentals. Actual data. I’d love to read what you can come up with.
Effed Orff retweeted
If you’re wondering why this selloff is so aggressive, save this! !
NASDAQ is down 5.1%. S&P 500 is down 3%. $MU −17%. $SNDK −18%.
This is what forced selling looks like.
It isn't happening in one sector. It's literally everywhere you look at once from financials to AI infrastructure and the best businesses like $MU $AMZN and $NVDA are falling hardest because they're the easiest things to sell when funds need cash…
It didn't have to be this way folks.
AI is the golden goose of this cycle…record earnings, contracted demand into 2030 and a market that was finally starting to price it.
But the war with Iran sent oil above $100, pushed inflation back up, forced the Fed into a hike and set the stage for last night's republican humiliation.
The cost of that war is now showing up in every portfolio in America including all the pension funds & 401Ks.
But what hasn't changed is the thesis.
3 days ago $AMD guided above everyone’s expectations.
$SMCI did $19.9bn in a quarter.
$MU and $SNDK are sold out for years, with margin floors written into their contracts. None of that has changed at all…in fact every earnings we get from here gets even better.
The fundamentals are intact. But every chart looks broken rn and has been broken for a couple of months.
These so called analysts spent this morning cutting their 2027–28 AI capex estimates because of the election.
The outcome had been telegraphed for months…these guys have no idea what they’re talking about…just as always chasing the news cycle.
Hyperscalers sign multi year capacity contracts and they don't rewrite them because control of the House changed hands.
What the market needs now is for the macro to stop getting worse.
Trump has accepted the results which takes a constitutional crisis off the table.
An Iran deal and lower oil would open the door for the Fed to cut. When this nonsense “war” fades, earnings are what's left and the earnings are still extraordinary and that’s where the opportunity is…
This is a flush mainly algo driven not the end of the AI trade or American capitalism...stay hard!
