Michael Fritzell (Asian Century Stocks)
This is the moment where everyone's convinced that we'll lose our jobs to AI tools:

"In their calmer moments, investors recognize their inability to know what the future holds. In moments of extreme panic or enthusiasm, however, they become remarkably bold in their predictions; they act as though uncertainty has vanished and the outcome is beyond doubt. Reality is abruptly transformed into that hypothetical future where the outcome is known. These are rare occasions, but they are also unforgettable: major tops and bottoms in markets are defined by this switch from doubt to certainty." - Peter Bernstein
tweet

God of Prompt
RT @godofprompt: this article exposes the gap between those that learn AI now and those who will choose to wait on the sidelines

the gap is widening. choose your side.

https://t.co/moCLsoNKwy - Alex Prompter
tweet

Benjamin Hernandez😎
$COST: Defending the $1,000 Level

Costco ($COST) surged +2.12% to end at $998. The battle for four digits is real. Investors are paying a premium for the membership moat as the broader market volatility spikes.

$ORCL $AMD $INTC $NVDA $TSM $TCNNF https://t.co/702oiL2Pj2
tweet

Jukan
People often say that optics will be the next big money-making field after memory, right?

This is an interesting article introducing noteworthy Japanese players in optics that you should pay attention to when considering investments in the sector.

Among the companies featured here, Anritsu (6754T) in particular caught my eye as especially intriguing.

https://t.co/LLKaGrwD1X - 슈포 (Superposition)
tweet

Quiver Quantitative
Senator Katie Britt was just asked about her stock trades.

She said she doesn’t trade any stock.

A screenshot from Quiver was shown, showing the trades she filed.

Then she pivots and says that proceeds from her stock trades go to charity.

https://t.co/r9t7V8biwx
tweet

Michael Fritzell (Asian Century Stocks)
The irony of SaaS bears setting up Substacks, giving away 10% of the economics, while open source alternatives like Ghost are readily available
tweet

Brady Long
The only thing that can kill your startup is you.

And it will most likely be because you spent too much time comparing yourself to the next guy or girl.
tweet

Michael Fritzell (Asian Century Stocks)
RT @CompoundingUp: Heian Ceremoney Service (2344) MBO at 1500 yen per share.

Shareholders on the Yahoo bulletin board feel it's too cheap.

I got in around 950 yen/share.

Before the announced MBO and recent spike, 2344 was trading at approx. -1.7bn EV, MCap 11bn yen, 8x PE, Net Cash/Investment securities of 13 bn, no debt, P/B 0.5x

The offer price sets it at 2.5x EV/EBIT, 13x PE, 0.8x P/B

I guess I owned this for a little over 1 month for +50%+ gain
tweet

Michael Fritzell (Asian Century Stocks)
RT @yieldsearcher: What if… the ultimate pro-AI trade is TLT.
tweet

Illiquid
You can have no AI exposure, just riding the Singapore government putting out bullish takes on construction. H/T to @Floebertus for finding this theme all the way in Europe. https://t.co/TdgLgvAX23
tweet

Jukan
Memory Big 3 HBM4 Competition Heats Up... NVIDIA Backing Off Its 11Gbps Insistence with Strategy Shift?

Competition among the global memory semiconductor Big 3 over HBM4 (6th-generation High Bandwidth Memory) supply is intensifying. Samsung Electronics announced HBM4 mass production shipments just the day before (Feb. 12), one step ahead of SK Hynix and Micron, emphasizing its development speed and performance advantages over competitors.

However, this year's HBM4 supply chain landscape is expected to be most heavily influenced by NVIDIA's HBM4 procurement strategy as the end customer. While HBM4 performance matters, timely launch of the latest AI accelerators requires consideration of HBM4 yield and supply stability as well. Indeed, the industry is paying close attention to the possibility that NVIDIA may relax its HBM4 performance requirements or accept next-tier products alongside top-spec ones.

Samsung Electronics Accelerates Shipment News... Power Struggle with SK Hynix and Micron

According to industry sources on Feb. 13, major memory companies including Samsung Electronics, SK Hynix, and Micron are set to proceed with risk production shipments of HBM4 for NVIDIA this month.

HBM4 is the next-generation HBM to be mounted on "Rubin," the AI accelerator that global Big Tech company NVIDIA plans to launch this year. Its key feature is doubling the number of I/O (input/output) pins—the data transmission pathways—to 2,048 compared to the previous generation.

Samsung Electronics was the first to officially announce HBM4 mass production shipments on Feb. 12. The company stated, "From the initial development stage of HBM4, we set performance targets exceeding JEDEC (Joint Electron Device Engineering Council) standards and pursued development accordingly," adding, "This product has secured stable yields and industry-leading performance from the early mass production stage without any redesign."

The industry is noting that Samsung Electronics announced its HBM4 mass production shipment news earlier than expected. Piecing together information from inside and outside Samsung, the company had originally planned to announce HBM4 mass production shipments on Feb. 19. However, as competitors have also recently made official announcements regarding HBM4 mass production, Samsung appears to have moved up its product shipment as much as possible to secure the initiative.

SK Hynix also plans to ship HBM4 products to NVIDIA this month. During its earnings conference call on Jan. 28, SK Hynix stated, "We are the only company in the industry capable of stably supplying both HBM3E and HBM4 simultaneously," adding, "In particular, HBM4, for which we established the industry's first mass production system in September last year, is currently in mass production for volumes requested by customers."

Micron also officially announced its shipment news on Feb. 11 (local time) during a brokerage conference call, directly refuting recent controversy over its alleged exclusion from the HBM4 supply chain. Micron stated, "HBM4 is already ramping up mass production, and we have begun shipping to customers," emphasizing, "This is approximately one quarter ahead of the guidance mentioned in our earnings call late last year."

All Three Memory Companies in Risk Production... Volume Expansion from H2 Onward

The three memory companies have recently been engaged in fierce jockeying for dominance over HBM4 commercialization. The industry is concerned that this process is generating unnecessary confusion and overinterpretation in the market.

Notably, the term "mass production" as used by the three memory companies differs from the conventional concept of mass production. In typical manufacturing, samples undergo quality testing with the customer, and once certain criteria are met, official purchase orders (POs) are placed.

In contrast, the current HBM4 efforts of all three memory companies are classified as "risk production." Risk production refers to the preemptive [...]

commitment of wafers for product manufacturing before customer certification is complete.

Shipping HBM as a final product requires four months for core die production alone. If mass production were to begin only after receiving official POs, NVIDIA as the end customer would be unable to launch its AI accelerators on schedule. Therefore, all three memory companies have been conducting production with the goal of shipping products this month to supply HBM4 to customers ahead of official POs.

A senior industry executive in the memory semiconductor sector explained, "NVIDIA's official qualification test completion date is set for the end of Q1 this year, after which official POs are expected to follow. While the memory companies are publicly using the term 'mass production,' strictly speaking, this falls under the concept of risk production."

Considering this, the point at which Samsung Electronics and SK Hynix's HBM4 shipment volumes begin to meaningfully expand is estimated to be H2 2025 at the earliest.

All Eyes on NVIDIA's HBM Procurement Strategy... Weighing Performance vs. Supply Stability

Currently, the prevailing industry view is that Samsung Electronics will complete NVIDIA's HBM4 qualification testing most smoothly. Samsung applied 1c (6th-generation 10nm-class) DRAM to its HBM4, which is one generation ahead of competitors. The base die that controls the HBM was also manufactured using the most advanced 4-nanometer (nm) process.

In its HBM4 mass production shipment materials, Samsung stated, "Our HBM4 has stably achieved an operating speed of 11.7Gbps, approximately 46% above the industry standard of 8Gbps, setting a new benchmark," adding, "This represents approximately a 1.22x improvement over the maximum pin speed of 9.6Gbps for the previous HBM3E, with capability to reach up to 13Gbps."

However, the industry is more focused on NVIDIA's product procurement strategy than HBM4 shipment speed. The concern is that if NVIDIA unconditionally insists on 11.7Gbps-class performance from memory suppliers, it will be unable to secure sufficient HBM4 volume needed for Rubin chip mass production.

First, Samsung Electronics lacks the capacity to rapidly scale up HBM4 shipments. Samsung's 1c DRAM yield is estimated at around 60% as of this month. Factoring in subsequent back-end processing, HBM yields drop even further. Additionally, 1c DRAM production capacity stood at approximately 60,000–70,000 wafers per month as of late last year—far short of what is needed to meet NVIDIA's total HBM4 demand. New and conversion investments in 1c DRAM are underway, but more time is needed before they are reflected in actual production capacity.

SK Hynix has been allocated the largest share of HBM4 volume (approximately 60% share) in its discussions with NVIDIA. However, it received assessments during initial HBM4 reliability evaluations indicating difficulty achieving 11Gbps-class performance. SK Hynix has been making hardware-level improvements to its HBM4 up until recently, but 100% success cannot be guaranteed.

For this reason, the industry expects NVIDIA to procure not only 11.7Gbps products but also next-tier products at 10.6Gbps and similar speeds. In this scenario, HBM4 supply would become considerably easier for all three memory companies from a technical standpoint.

An industry source explained, "The HBM supply chain must consider not just performance alone, but power efficiency, thermal management, cost, and supply stability all at once. Amid a memory shortage incomparably worse than last year, it is virtually a foregone conclusion that NVIDIA will relax its HBM4 performance requirements to ensure supply stability."

$MU
tweet

Jukan
I previously told you—before anyone else—that NVIDIA had lowered its 11Gbps requirement.

-------------------------------
'An industry source explained, "The HBM supply chain must consider not just performance alone, but power efficiency, thermal management, cost, and supply stability all at once. Amid a memory shortage incomparably worse than last year, it is virtually a foregone conclusion that NVIDIA will relax its HBM4 performance requirements to ensure supply stability."'

Memory Big 3 HBM4 Competition Heats Up... NVIDIA Backing Off Its 11Gbps Insistence with Strategy Shift?

Competition among the global memory semiconductor Big 3 over HBM4 (6th-generation High Bandwidth Memory) supply is intensifying. Samsung Electronics announced HBM4 mass production shipments just the day before (Feb. 12), one step ahead of SK Hynix and Micron, emphasizing its development speed and performance advantages over competitors.

However, this year's HBM4 supply chain landscape is expected to be most heavily influenced by NVIDIA's HBM4 procurement strategy as the end customer. While HBM4 performance matters, timely launch of the latest AI accelerators requires consideration of HBM4 yield and supply stability as well. Indeed, the industry is paying close attention to the possibility that NVIDIA may relax its HBM4 performance requirements or accept next-tier products alongside top-spec ones.

Samsung Electronics Accelerates Shipment News... Power Struggle with SK Hynix and Micron

According to industry sources on Feb. 13, major memory companies including Samsung Electronics, SK Hynix, and Micron are set to proceed with risk production shipments of HBM4 for NVIDIA this month.

HBM4 is the next-generation HBM to be mounted on "Rubin," the AI accelerator that global Big Tech company NVIDIA plans to launch this year. Its key feature is doubling the number of I/O (input/output) pins—the data transmission pathways—to 2,048 compared to the previous generation.

Samsung Electronics was the first to officially announce HBM4 mass production shipments on Feb. 12. The company stated, "From the initial development stage of HBM4, we set performance targets exceeding JEDEC (Joint Electron Device Engineering Council) standards and pursued development accordingly," adding, "This product has secured stable yields and industry-leading performance from the early mass production stage without any redesign."

The industry is noting that Samsung Electronics announced its HBM4 mass production shipment news earlier than expected. Piecing together information from inside and outside Samsung, the company had originally planned to announce HBM4 mass production shipments on Feb. 19. However, as competitors have also recently made official announcements regarding HBM4 mass production, Samsung appears to have moved up its product shipment as much as possible to secure the initiative.

SK Hynix also plans to ship HBM4 products to NVIDIA this month. During its earnings conference call on Jan. 28, SK Hynix stated, "We are the only company in the industry capable of stably supplying both HBM3E and HBM4 simultaneously," adding, "In particular, HBM4, for which we established the industry's first mass production system in September last year, is currently in mass production for volumes requested by customers."

Micron also officially announced its shipment news on Feb. 11 (local time) during a brokerage conference call, directly refuting recent controversy over its alleged exclusion from the HBM4 supply chain. Micron stated, "HBM4 is already ramping up mass production, and we have begun shipping to customers," emphasizing, "This is approximately one quarter ahead of the guidance mentioned in our earnings call late last year."

All Three Memory Companies in Risk Production... Volume Expansion from H2 Onward

The three memory companies have recently been engaged in fierce jockeying for dominance over HBM4 commercialization. The industry is concerned that this process is generating unnec[...]

essary confusion and overinterpretation in the market.

Notably, the term "mass production" as used by the three memory companies differs from the conventional concept of mass production. In typical manufacturing, samples undergo quality testing with the customer, and once certain criteria are met, official purchase orders (POs) are placed.

In contrast, the current HBM4 efforts of all three memory companies are classified as "risk production." Risk production refers to the preemptive commitment of wafers for product manufacturing before customer certification is complete.

Shipping HBM as a final product requires four months for core die production alone. If mass production were to begin only after receiving official POs, NVIDIA as the end customer would be unable to launch its AI accelerators on schedule. Therefore, all three memory companies have been conducting production with the goal of shipping products this month to supply HBM4 to customers ahead of official POs.

A senior industry executive in the memory semiconductor sector explained, "NVIDIA's official qualification test completion date is set for the end of Q1 this year, after which official POs are expected to follow. While the memory companies are publicly using the term 'mass production,' strictly speaking, this falls under the concept of risk production."

Considering this, the point at which Samsung Electronics and SK Hynix's HBM4 shipment volumes begin to meaningfully expand is estimated to be H2 2025 at the earliest.

All Eyes on NVIDIA's HBM Procurement Strategy... Weighing Performance vs. Supply Stability

Currently, the prevailing industry view is that Samsung Electronics will complete NVIDIA's HBM4 qualification testing most smoothly. Samsung applied 1c (6th-generation 10nm-class) DRAM to its HBM4, which is one generation ahead of competitors. The base die that controls the HBM was also manufactured using the most advanced 4-nanometer (nm) process.

In its HBM4 mass production shipment materials, Samsung stated, "Our HBM4 has stably achieved an operating speed of 11.7Gbps, approximately 46% above the industry standard of 8Gbps, setting a new benchmark," adding, "This represents approximately a 1.22x improvement over the maximum pin speed of 9.6Gbps for the previous HBM3E, with capability to reach up to 13Gbps."

However, the industry is more focused on NVIDIA's product procurement strategy than HBM4 shipment speed. The concern is that if NVIDIA unconditionally insists on 11.7Gbps-class performance from memory suppliers, it will be unable to secure sufficient HBM4 volume needed for Rubin chip mass production.

First, Samsung Electronics lacks the capacity to rapidly scale up HBM4 shipments. Samsung's 1c DRAM yield is estimated at around 60% as of this month. Factoring in subsequent back-end processing, HBM yields drop even further. Additionally, 1c DRAM production capacity stood at approximately 60,000–70,000 wafers per month as of late last year—far short of what is needed to meet NVIDIA's total HBM4 demand. New and conversion investments in 1c DRAM are underway, but more time is needed before they are reflected in actual production capacity.

SK Hynix has been allocated the largest share of HBM4 volume (approximately 60% share) in its discussions with NVIDIA. However, it received assessments during initial HBM4 reliability evaluations indicating difficulty achieving 11Gbps-class performance. SK Hynix has been making hardware-level improvements to its HBM4 up until recently, but 100% success cannot be guaranteed.

For this reason, the industry expects NVIDIA to procure not only 11.7Gbps products but also next-tier products at 10.6Gbps and similar speeds. In this scenario, HBM4 supply would become considerably easier for all three memory companies from a technical standpoint.

An industry source explained, "The HBM supply chain must consider not just performance alone, but power efficiency, thermal management, cost, and supply stability all at once. Amid a memory shortage incomp[...]

arably worse than last year, it is virtually a foregone conclusion that NVIDIA will relax its HBM4 performance requirements to ensure supply stability."

$MU - Jukan tweet