DanielRomero

Morgan Stanley says the setup for legacy memory is getting tighter than expected
DDR4 pricing could rise another 20% in Q3 as supply keeps shrinking, with the bank now modeling a 19%-20% supply-demand gap in 2H26 and 18%-20% gaps in 2027-2028
Other industry coverage:
TrendForce says DDR4 will remain in short supply through 2026, even after Micron’s Virginia Fab 6 ramp
$MU move is mostly an internal capacity reallocation, not a real increase in total LPDDR4/DDR4 capacity. TrendForce expects DDR4 to represent only around 7% of Micron’s DRAM output in 2026, with that share declining further in th

Samsung started shipping HBM4E samples to customers
The chip is over 20% faster than its previous HBM4 product and uses Samsung’s latest 1c DRAM, plus its own 4nm foundry base die
Anthropic also named Samsung as a strategic infrastructure partner, and Samsung was highlighted for its logic chip capabilities, not just memory
Current HBM share as of Q4 2025:
- > SK hynix: 57%
- > Samsung: 22%
- > $MU: 21%

$10B over 5 years for quantum computing investments seems like a major conviction leap for the industry
$IBM generates around $15B in FCF per year. Dedicating roughly 13% of annual FCF to this new technology is massive

$DELL said its customer count surpassed 5,000, with growth across neocloud, sovereign, and enterprise customers
This was a fairly bullish call for $PENG
Trading at 26x forward P/E, with revenue growth accelerating
Not cheap, but the next prints should be good

$DELL +38% now in after-hours
$NVDA is pumping every company in the supply chain
Jensen single-handedly created millions of new millionaires

As mocked as it is, the bottleneck trade will stay forever
You only need to project the next 100 years to realize that AI is going to bottleneck everything
The first bottlenecks appeared in the areas most directly exposed. That was memory and power access
Then it became substrates, MLCCs, logic fabs, lasers, even copper foil and power semis
It’s only a matter of time before AI and robotics/physical AI accelerate the rate of progress to the point where every physical asset becomes shortaged
Maybe I’m wrong, and this is clearly a very early thought, but the everything shortage is a natural conse

Many "bottleneck" stocks pitched on X will, at some point, go down 90%
I’m very happy with the theme because I’m heavily exposed to it, but I hope everyone realizes this situation is not normal
There are stocks exploding for the right reasons, but that’s not the case for many

Are we going to pretend this Iran-induced dip won’t get bought and that we won’t be back at ATHs (for AI buildout trades) in a week again?
The only thing that can stop this momentum is rampant inflation that forces the Fed to raise rates
Unless that happens (which would directly affect capex plans) every dip is a buying opportunity

Did Leopold Aschenbrenner just put 20% of his fund into $NBIS?
~$14B AUM, and he just took a 5.6% stake in $NBIS
5.6% of $NBIS ~$53B market cap is $3B
$3B / $14B = 21%
I was expecting him to take a position at some point, but becoming one of the largest shareholders is something else

$NVDA Vera CPU tops $AMD and Intel in its first independent benchmarks
NVIDIA’s Vera CPU has posted its first independent benchmarks, with Phoronix showing the 88-core Arm processor delivering the best performance ever seen from an Arm chip, including a 10% advantage over AMD’s high-frequency EPYC 9575F on a geometric mean basis
Vera uses 88 custom Olympus cores built on Armv9.2-A, supporting 176 threads through NVIDIA’s Spatial Multithreading. Unlike Grace, which used off-the-shelf Arm Neoverse cores, Vera is a ground-up custom NVIDIA design that the company claims delivers roughly double Gra

$QCOM 必須是今年最令人意想不到的爆炸性贏家
它五年什麼都沒做，然後在一個月內創下歷史新高

We’re living history here
This is the bottleneck era
I’ve never seen a market so driven not by what’s exciting, but by what’s scarce
There have always been bottlenecks, but this year has been insane, almost like a social experiment

$TSM 3nm chips are expected to see a 15% price increase in the second half of this year, with another 5% to 10% hike possible next year
TSMC’s Fab 18 is reportedly running at high utilization, with 3nm monthly capacity rising from around 130,000 wafers at the start of the year to roughly 160,000 to 175,000 wafers in Q2. Even with this ramp, AI demand continues to exceed expectations
At the same time, MSCI’s semi annual adjustment takes effect after market close on the 29th, with TSMC receiving the largest weighting increase in the MSCI Taiwan Index, rising 0.56 points to 58.33%. Foreign invest

$NVDA next-generation Vera Rubin AI platform is expected to significantly increase the content value of the power, cooling, and optical communication supply chain
Compared with Blackwell, Vera Rubin’s power consumption doubles, while performance per watt improves 10x
This means AI racks will consume much more power, pushing power supply requirements higher and accelerating demand for 800V HVDC architectures
The main beneficiaries are:
Power supply chain
Delta Electronics ( and Lite-On ( are expected to benefit from higher-end power supply requirements as AI racks move toward higher voltage, hi

$NVDA Rubin is driving a major hardware content upgrade across the AI server supply chain, according to Morgan Stanley’s latest Taiwan supply-chain analysis
The biggest beneficiaries are:
1. PCBs
PCB content value is expected to rise 233% versus GB300
Morgan Stanley estimates Rubin PCB content at around $117,000 per rack, compared with roughly $35,000 for GB300
2. Passive components
Passive component content value is expected to rise 182%, mainly due to higher computing density, power density, and system complexity
For MLCCs specifically, Morgan Stanley estimates VR200 architecture content val