Oesta

[Oesta @OestaOne]

@HighyieldHarry "Incredible win for non-AI writing."

You underestimate the ability of AI to perfectly simulate human writing if it wants to.

[Oesta @OestaOne]

@babyfolio

[Oesta @OestaOne]

@naval Not really. Working smart can sometimes remove the need for hard work, especially in the age of increasing technological leverage like AI.

[Oesta @OestaOne]

@J5jayfive First half? Scaling of agentic AI has barely begun.

[Oesta @OestaOne]

@Brian_Stoffel_

[Oesta @OestaOne]

@pandawatch88 Either way that means the US gets lesser memory allocation.

Incentivising memory makers to sell to others at full price vs selling to US at a discount means US will get lesser of AI's most strategic component.

Losing the most important technological race just for a discount.

[Oesta @OestaOne]

@oguzerkan Low IQ take.

Samsung's business that are negatively affected by memory is basically negligable in terms of overall figure percentage, even if you discount its mobile phone business as a zero it doesn't make a dent.

[Oesta @OestaOne]

@jukan05 Who cares if the earning yield is so high.

[Oesta @OestaOne]

@firstadopter Earning more operating profit than the world's most valuable company isn't enough.

[Oesta @OestaOne]

@MarcosMillaYT Incorrect.

https://t.co/ivR0VBY7zj

[Oesta @OestaOne]

@MarcosMillaYT It's actually Samsung.

1. Its baseline is much higher than the other 2. P/B ratio also much lower.

2. Diversified business including world's 2nd best logic fab (free optionality).

3. Cheapest valuation in terms of forward P/E.

[Oesta @OestaOne]

@SemiconductorsX IPO hype talk

[Oesta @OestaOne]

@AlmaCap114204 Just common sense not to fully rely the most critical technology on 1 company and 1 country, especially the geopolitical hotspot.

[Oesta @OestaOne]

@Blinklebloop And all of it need memory to function.

[Oesta @OestaOne]

@rubicon59 People talking about "effiency improvements" of memory like it isn't happening to compute.

Efficiency improvements has been happening on compute and memory since its invention, yet their demand has never been higher.

[Oesta @OestaOne]

@brandonjcarl Don't think you understand all of these points are bullish.

Companies don't enter or expand capacity if the outlook is poor.

And these companies have far more industry visibility than almost anyone else, especially you.

[Oesta @OestaOne]

@MikeFritzell @joinyellowbrick https://t.co/ZGBd5jwpy1

[SemiVision 🇹🇼 👁️👁️ @semivision_tw]

AI Chips Are Hard. Memory Is Harder.

https://t.co/6e0E7kKbaR

[Oesta @OestaOne]

@MikeFritzell @joinyellowbrick Facts.

https://t.co/Y09Plb1CDi

[SemiAnalysis @SemiAnalysis_]

A good chunk of inference for the most successful AI agent, Claude Code, is done on Trainium, while Claude training is done on TPUs. Just a year ago, it would have been unimaginable that TPUs and Trainium could grow this rapidly, while the CUDA moat slowly eroded. 3/4🧵

[Oesta @OestaOne]

@MikeFritzell @joinyellowbrick CUDA is software moat. Memory is complex manufacturing moat.

No prizes for guessing which is the stronger and sustainable moat in the AI era.

[Oesta @OestaOne]

@joinyellowbrick And BTW, Nvidia is much easier to replace than Memory.

Anyone can design AI accelerators. Now try building a memory fab.

[Oesta @OestaOne]

@joinyellowbrick People said the same about Nvidia 3 years ago.