7
https://x.com/i/lists/1669153613199835138?t=R0mCicxs7zfJE_yOAek4gQ&s=09
Jukan
Why Did Samsung Only Receive 20% of Nvidia's HBM4 Allocation?- Hankyung
Samsung Electronics is set to begin mass production and shipment of HBM4 (6th-generation High Bandwidth Memory) to Nvidia this month — an industry first. However, Samsung's share of Nvidia's total HBM4 demand is reportedly only in the "mid-20%" range. Samsung's HBM4 boasts the industry's highest performance (operating speed of 11.7 gigabits per second) and was the first to pass the quality (qual) test. So why is Samsung's projected share of Nvidia's allocation still behind SK hynix (mid-50% range), which is still undergoing its final qual test? We sought to answer this question that has been growing in the market recently.
According to semiconductor industry sources on the 9th, Nvidia tentatively allocated HBM4 volumes to Samsung Electronics, SK hynix, and Micron in December of last year — SK hynix at mid-50%, Samsung at mid-20%, and Micron at around 20%. There is a reason Nvidia allocated volumes before qual tests were even completed: since HBM production takes "more than six months," volumes must be assigned to suppliers in advance to ensure stable production of Nvidia's new AI accelerator, Vera Rubin, equipped with HBM4, in the second half of this year. The allocations reportedly took into account years of transaction history with HBM suppliers, each company's HBM4 production capacity, and the likelihood of passing qual tests.
Satisfied with Proving HBM4 Technical Prowess… Samsung's Profitability Maximization Strategy
Samsung's 20%-range share has drawn "disappointing" assessments. But there are reasons behind it. First, there is Samsung's own strategy. Samsung poured everything into proving its competitiveness in the HBM4 market, which is ramping up this year — essentially an "all-in on technology" strategy. This is exemplified by the use of 4nm foundry technology for the base die (the "brain" of HBM4) and 10nm 6th-generation (1c) DRAM for the core die (the basic building block).
While the 4nm base die is not the most cutting-edge like 2nm, it still belongs to the advanced foundry tier. And 1c DRAM is currently the most advanced DRAM product available. Compared to a competitor using an older 12nm foundry process with 10nm 5th-generation (1b) DRAM, Samsung's HBM4 inevitably has higher production costs.
Yield is another factor. While there have been recent reports in the semiconductor industry that yields for HBM4's 1c DRAM have improved significantly, they still fall short of a competitor's 1b DRAM, which has achieved higher process maturity. Depending on contract terms with customers (wafer-level delivery vs. individual chip-level delivery), the impact varies, but generally, lower yields mean lower profitability.
Production capacity is also still insufficient. The production capacity for 1c DRAM — the fundamental building block of HBM4 — is at approximately 70,000 wafers per month, which is about 10% of Samsung's total DRAM capacity. Although the company has recently begun expanding its Pyeongtaek Plant 4, production capacity won't increase to around 190,000 wafers until a year from now. Manufacturing HBM4 also requires "cutting-edge packaging" that makes the GPU and HBM function as a single chip. This process further reduces the number of surviving chips. From Samsung's perspective, even if it wanted to produce more HBM4, it simply cannot at this point.
Commodity DRAM Prices Have Risen to HBM3E Levels… Even Higher Profitability
Demand remains strong for HBM3E 12-layer products, which will serve as the market's mainstream through the first half of this year. Samsung passed Nvidia's HBM3E 12-layer qual test around September–October of last year, but actual shipment volumes are reportedly not large. Instead, Samsung is reportedly receiving a flood of orders for "HBM3E 12-layer" from Broadcom, which co-designs Google's AI accelerator TPU, among others. For Samsung, there is little incentive to cut production capacity for HBM3E 12-layer — which uses 10nm 4th-genera[...]
Jukan
* Korea Economic Daily: NVIDIA allocated the HBM4 volume it needs to the three memory makers last December, with the split at mid-to-high 50% for SK hynix, mid-20% for Samsung Electronics, and around 20% for Micron.
** This report runs contrary to SemiAnalysis’s estimates.
tweet
Michael Fritzell (Asian Century Stocks)
RT @TheStalwart: If you're looking for something to read, here's every book that's been recommended in the Odd Lots Discord https://t.co/0C7fIhYMtc
tweet
g your life
if you want to follow this path you need to adopt the rbi process which stands for research backtest and implement. you find the strategy in a book like market wizards and then you build a tool to see if the data supports the thesis. if the backtest looks good you put it into production and let the machine execute the vision without hesitation
the ultimate goal is to bridge the gap between social sentiment and market reality before the rest of the world catches on. it is about finding the starbucks limited cups or the next anti aging skincare line that is flying off the shelves. when you have an ai agent doing the observation for you the only thing left to do is make the decision
coding gave me the freedom to stop staring at charts and start thinking about the bigger picture of the markets. i do not put it past this technology to create the next generation of billionaires who never even worked on a trading floor. all it takes is the willingness to learn and the patience to iterate until the system works
the world is changing fast and the old ways of trading are becoming obsolete for the individual investor. you can either be the one who uses code to level the playing field or you can continue to be the liquidity for those who do. the choice is yours but the bots are already running and they are not waiting for anyone to catch up
tweet
Michael Fritzell (Asian Century Stocks)
RT @capitalemployed: If you have access to Thai stocks then this is an interesting idea - Carabao $CBG.BK 🇹🇭
New deep-dive on the website: on energy drink maker Carabao (CBG TB):
- Forward P/E 12x
- 70% decline in the share price since 2021
- Founder bought 1.25 million shares in June 2025 https://t.co/ej0ZGQ92jY
tweet
Illiquid
Don’t own enough of Japan. Even the umami company is on fire. https://t.co/scDrwp9h62
tweet
Michael Fritzell (Asian Century Stocks)
Now that SaaS is out of fashion, perhaps I'll finally be able to buy Smaregi 4431 JP at a reasonable price https://t.co/gXStYPqGD0
tweet
God of Prompt
RT @godofprompt: Do you really need a Claude Max subscription?
yes yes yes yes
yesyes yes yes yes
yes yes yes yes yes
yes yesyes yes yes
yes yesye yes yes
yes yes yes yes
tweet
Michael Fritzell (Asian Century Stocks)
Japan really is a two-tier market… I feel more comfortable with the growth index at this point
LOL, no love for Growth Index even as Nikkei surges. In fact all of my stocks gapped up and reversed all gains. https://t.co/V9CzqdZVe8
tweet
The Transcript
$INTC CEO on the key constraint:
"I think, for a lot of my customer is memory. Memory, actually, there's no relief as far as I know, when I talk to only 3 key players, 2 of them I talked to very frequently. And then they told me that Lip-Bu, there's no relief until 2028"
tweet
Javier Blas
Global electricity demand will increase at an annual rate of 3.6% from 2026 to 2030, the @IEA says in new forecast. Importantly, "although coal generation is set to lose ground globally, it remains the single largest source of electricity in 2030."
https://t.co/7kxaUjO0RZ
tweet
Michael Fritzell (Asian Century Stocks)
Key point from that Bloomberg article about Minal Bathwal: read Kahneman's book https://t.co/31AlkJvzCB
tweet
Illiquid
At the APE conference to see Terahop. 😂 https://t.co/YF9fFMDHVu
tweet
Michael Fritzell (Asian Century Stocks)
There’s been proliferation of AI generated content on Substack, too, but it tends not to be very good
The past year has seen an explosion in coding productivity @FT https://t.co/XhcrqLkxzv
tweet
Illiquid
I expect to find only a handful of top tier ideas every year. In the last five months, I found the following companies within 30 miles of where I live: a diesel genset company, a cable manufacturer, two builders, and Wasion. Will publish a post on my new favourite idea next week. The strongest idea yet I think, which is saying alot because Kwan Yong Holdings is still so cheap.
tweet
Brady Long
RT @thisguyknowsai: 10 Powerful Gemini 3.0 prompts that will help you build a million dollar business (steal them): https://t.co/oL0BVzPIum
tweet
Michael Fritzell (Asian Century Stocks)
After Coupang's leak, the entire company is on the verge of breaking down. In Substack's case, no-one is batting an eye?
Very strange - it's as if everyone's just taking cues from others on when and why they should become outraged
What's missing:
- A press release
- How many contact details were leaked (700,000)
- What else was leaked other than emails/phone numbers: specifically: 1) full names 2) user ID 3) Stripe ID 4) profile pictures 5) account creation dates 6) social media handles
- Why it happened
tweet
Michael Fritzell (Asian Century Stocks)
RT @HiddenMonopoly: Nice study !
High gross margins and a low volatility in gross margins may point to industries with high pricing power. https://t.co/koxTkBoLLt
tweet
Moon Dev
From $20K To $40M: The Python Strategy That Automates Social Arbitrage And Beats Bloomberg
most people are looking for the next big trade in a terminal but the real billion dollar use case for ai is hiding in the brain rot of social media. if you can catch a trend six months before the suits on wall street even hear about it you have an edge that no algorithm can beat
i believe code is the great equalizer because it allows one person to do the work of a thousand analysts. i spent hundreds of thousands on developers in the past because i thought i could not code myself but then i realized that liquidations were my only other option. now i build these systems live and let them work for me twenty four seven while i sleep comfortably knowing i am not over trading
there is a guy named chris who turned twenty thousand dollars into forty million by simply watching what was trending on social media before it ever hit a bloomberg terminal. he is featured in the market wizards books which are the gold standard for verifying the best traders in the world. the problem is that sitting on tiktok all day trying to find the next abercrombie or lululemon is exhausting for a human mind
this is where clawdbot comes into play to handle the heavy lifting of social arbitrage. she scans the platform every five minutes looking for specific signals that indicate a massive shift in consumer behavior. most traders are busy drawing lines on a chart while we are busy identifying products that are literally sold out everywhere
the secret sauce is in the keywords we use to filter through the noise of the internet. if you just look at your recommended feed you are just consuming content but if you search for terms like restock alert or tiktok made me buy you are looking at data. these are the breadcrumbs left by the next billion dollar brand before the stock price reflects the reality
wall street is comprised of an older generation that simply does not understand the speed of digital trends. they find out about things when the data hits their official reports which is usually months after the product has already gone viral. by the time they are buying we are already looking for the next exit or the next play
i recently updated the system to version one because version zero was looking at popular videos that might have been months old already. now the bot only focuses on videos uploaded within the last seven days to ensure the signal is as fresh as possible. this prevents us from chasing trends that have already peaked and allows us to get in while the fire is still just a spark
one example that popped up recently was the walmart birkin which is something i would never have thought about in a million years. it had hundreds of thousands of likes and was driving massive traffic to a specific product line. being a quant is about being a data dog and following where the numbers lead regardless of whether you personally understand the product
the reason i share all of this is because i want to see more people escape the trap of manual trading and liquidations. you have to iterate to success and that means testing a hundred ideas just to find the one that actually has legs. for every hundred brands the bot finds there might only be one that is worth a deep dive into the public company financials
most people will never take the time to learn how to code because they think it is too hard or too expensive. i am proof that you can start from zero and build fully automated systems that trade for you instead of against you. once you have the bot watching the market twenty four seven you have successfully removed the emotional baggage that causes most traders to fail
there is a massive shift happening where one person can run a billion dollar company using ai agents as their staff. sam altman has talked about this and i tend to agree because these bots do not get tired or bored. they will continue to scour the web for social arbitrage opportunities while you are out livin[...]
God of Prompt
RT @godofprompt: https://t.co/voGmuqhGiD
tweet
Michael Fritzell (Asian Century Stocks)
RT @atelicinvest: It's really simple, really.
All you really need to do is:
- get a clear understanding of LLM capabilities today
- make a correct bet on where that is heading within the next 1-3 years
- make a correct bet on customer behaviour / demand pattern
- make a correct bet on competitive behaviour - both existing and new
- make a correct bet on how well the management will handle the transition
tweet
God of Prompt
RT @godofprompt: Stop telling LLMs like Claude and ChatGPT what to do.
Start asking them questions instead.
I replaced all my instruction prompts with question prompts.
Output quality: 6.2/10 → 9.1/10
This is called "Socratic prompting" and here's how it works: https://t.co/UWgQXYHS0D
tweet
Jukan
I would like to express my gratitude to Prime Minister Takaichi. https://t.co/96GtQV6EmZ
tweet
Brady Long
RT @thisdudelikesAI: Love the remastering of the classic 1984 vibe but with robots marching through tubes instead of humans. Pretty sick tbh.
The real mind-blower tho is that AITOPIA’s SuperAgent generated the whole video and most of their store is AI-made already.
2026 AI Remastered
On February 24, 2026
AITOPIA will introduce AI App Store and a new AI economy: AIconomy
And you’ll see why 2026 won’t be like “2026.”
~84% of our AI App Store is created by AI
~100% of this video is created by AITOPIA SuperAgent
❤️🍎 🏈 https://t.co/yzfKL064Jr
tweet
Jukan
"A real trader should've seen Takaichi's landslide victory coming from a mile away and already had Nikkei longs loaded up in the portfolio, bro." https://t.co/nKvyBbFNqn
tweet
App Economy Insights
RT @EconomyApp: $AMZN Amazon Q4 FY25:
• Revenue +14% Y/Y to $213.4B ($2.2B beat).
• Operating margin 12% (+0.4pp Y/Y).
• EPS $1.95 ($0.01 miss).
• Q1 FY26 rev guide: +11%–15% Y/Y.
☁️ AWS:
• Revenue +24% Y/Y to $35.6B.
• Operating margin 35% (-2pp Y/Y). https://t.co/169cUDweIt
tweet
Jukan
* Reuters: Intel told customers in China that lead times for its server CPUs could be as long as six months.
* Reuters: Intel raised prices for server CPUs sold to China by 10%.
* Reuters: AMD said lead times for some of its products have extended to 8–10 weeks.
$AMD $INTC https://t.co/T0i9NrwXNH
tweet
The Transcript
RT @TheTranscript_: Alphabet CBO: "For FY 2025, our YouTube's annual revenue surpassed $60B across ads and subscriptions. In Q4, YouTube ads was driven, indeed, by strong growth and direct response." $GOOG $GOOGL https://t.co/oB5gWSIFmX
tweet
Jukan
UBS raised its price target on Micron to $450 with the following commentary...
"More importantly, synthesizing industry conversations suggests that memory supply shortages could extend deeper through C2027 — DRAM could remain in a supply deficit through C4Q27, and NAND could be in shortage through C1Q27.
This is occurring despite our view that accelerating memory price inflation will cause some demand destruction and spec-downs in PCs and smartphones (accordingly, we have lowered related estimates). The key takeaway is that strong data center demand is more than offsetting these headwinds.
(omitted)...
We are modeling gross margins of 69.9%, with an increase sequentially in the May quarter (MayQ) to approximately 74.4%. Accordingly, we expect gross margins to peak in the low 80% range across C1Q27E/C2Q27E."
$MU
tweet
Jukan
The Big 3 memory companies are basically extorting money at this point. They're refusing to honor the prices from previously signed contracts and are pressuring people to pay according to current market prices lol https://t.co/7ikDqMCuSx
Memory Big 3 Change Contract Practices… Post-Settlement Pricing Emerges
Rising memory prices are now transforming supply contract practices. Contract durations are shifting from long-term to short-term, and a "post-settlement" concept reflecting market prices has emerged, moving away from negotiated fixed pricing.
Based on reporting by ETNews on the 4th, Samsung Electronics, SK hynix, and Micron — the major memory manufacturers — have recently been entering new types of product supply contracts. An industry insider familiar with the matter said, "As memory prices have surged, an unprecedented post-settlement pricing method has emerged, and ultra-short-term contracts are also being adopted."
Typically, memory products such as DRAM and NAND have prices specified at the time of the initial supply contract. With fixed transaction prices, even if market conditions change and prices rise or fall, adjustments are generally made within approximately 10% through negotiations. For example, if a contract is signed to supply DRAM at 100 won, the price mostly holds for about one year. If there are price fluctuations, quarterly negotiations within the contract period adjust it to 110 won (+10%) or 90 won (-10%) for supply in the following quarter.
However, recently, contracts have been emerging where price increases are compensated retroactively to reflect market rates even after supply has been delivered. In other words, even if a contract was signed to supply DRAM at 100 won for one year, if DRAM market prices have risen 100% by the end of the contract period, an additional 100 won is paid. This is essentially a post-settlement arrangement.
All three major memory companies have already signed contracts using this method, and the counterparties are primarily North American big tech clients. While this could be disadvantageous for suppliers if memory prices fall, analysts note that the risk is minimal since price increases are currently expected to far outweigh the risk of decline.
Another industry insider explained, "From major clients' perspective, securing memory supply is currently more important than the contract format," adding, "They have determined that it is urgent to lock in supply contracts first, even if additional costs arise later."
Contract durations are also changing. Memory clients want long-term contracts of two years or more, beyond the typical one year, to secure stable memory supply for the expansion of artificial intelligence (AI) infrastructure. However, memory manufacturers as suppliers are increasingly reluctant to agree to this. Supply is tight and price volatility is high. There is a risk that locking into a long-term supply contract with one client could mean missing opportunities to secure other clients on better terms.
As a result, contracts have reportedly shifted from annual to quarterly, and even monthly terms. In one known case, a North American data center operator A requested a two-year long-term supply contract from one memory manufacturer but was rejected, and barely managed to secure a supply commitment from another manufacturer. This contract also included the post-settlement pricing method for price increases.
An industry insider said, "This supplier-favorable contracting trend is expected to continue at least until the second half of the year, when the upward momentum in memory prices is expected to moderate."
tweet