사과나무

Invester

[사과나무 @treeapple33]

@jukan05 저런 얘기가 나올만큼 미국이 앞서있는 건 아닌 것 처럼보이는데, 굉장한 패착이 될지도 모른다고 생각.. 🤔

[사과나무 @treeapple33]

믿어도 되는걸까...? 🤔

[Alexandr Wang @alexandr_wang]

1/ we’ve built a lot of connectors to make it easier to integrate Muse into your life. Muse runs in a secure VM and we don’t use your data for ads or anything other than to make Muse great for you. you can connect Muse to Gmail, Google calendar, Outlook, Plaid, Opentable, Google Docs, Spotify, Function Health, Withings, Tailscale, Peloton, and more.

[Steven Heidel @stevenheidel]

token pricing is effectively meaningless now. 3.8 flash looks 13x cheaper when measured per token, but Astra is cheaper per task since it's far more efficient. measure your costs per task, not per token.

[사과나무 @treeapple33]

@jukan05 흠 사실 뻔한 대형기업들 제외하면 인텍정도 보고 있습니다. 지난 사이클 때 봤던 기업이라 다시 체크 중인데 여전히 인텔 쪽은 Large formfactor독점인 것 같네요. 패키지 검사라 좀 간접적인 것도 같고.

[사과나무 @treeapple33]

@jukan05 이번 사이클이 끝날쯔음 일본 밸류체인들이 힘을 많이 잃을 것도 같아요. 대만과 중국에 의해

[Ryan Wang 🇹🇼 @ryanwang]

HBM splits into two things: capacity (how many layers) and speed (bandwidth/latency). Capacity decides whether KV, batch, and context fit on the card. Speed decides how many bytes you can move each step. Stacking mainly adds GB. It does little for bandwidth and can make the cube hotter, slower, and harder to supply.

The old script was 12/16-Hi and ever-larger memory per GPU. Rubin Ultra rumors now point to 8-Hi / 192GB. At Hot Chips, some labs were asking for 4-Hi. Same direction, different notch: not “no HBM,” but no paying heat and yield for unused capacity.

For inference and agents, the scoreboard is tokens per watt, then tokens per dollar. Capacity is the entry ticket. Bandwidth is the throttle. “4-Hi is enough” means: keep pin speed, make up capacity with NVLink and the rack, and a shorter stack can turn watts into tokens more cleanly. That pushes out the urgency of very tall stacks and hybrid bonding. Flagship training SKUs may still keep high-capacity parts. Do not read this as HBM demand disappearing.

[Mark Kretschmann @mark_k]

Cloud agents are the future. Say goodbye to the harness running locally on your computer.

All models will move toward agent swarms and cloud computers very soon, much like Grok Bot.

[사과나무 @treeapple33]

Grok bot이 next claude code가 될 듯..

[사과나무 @treeapple33]

@zephyr_z9 I wonder what they're looking at, or if they haven't seen it.

[사과나무 @treeapple33]

@jukan05 저도 정확히 이 생각으로 광학 그리고 기판 쪽으로 포트폴리오를 완전히 돌렸습니다. 좀 많이 아쉽긴하지만..
근데 중장기적으로봐도 솔직히 메모리의 상방이 썩 매력적으로 보이진않는 것 같아요 ㅋㅋ 오히려 메모리 트레이드가 안정되면 WFE가 나을지도..

[사과나무 @treeapple33]

@pequityresearch However, they couldn't give dopamine to the stock market player 😂

[Susan Zhang @suchenzang]

backdrop: kings all asleep/on vacation while minions slave away

august 2024 - first big fail

jan 2025 - only competent vp bails after one terrorist lead demanding an even bigger slice of pie breaks the camels back

summer 2025 - second big fail + the king of terrorists climbs the ranks above the slumbering execs

summer 2026 - slumbering execs + terrorists who only show up for successful announcements move on, leaving just the king of terrorists behind

-----

lesson: if you're high enough up there, you can pretty much be a terrorist and hold orgs hostage while continuing to ask for more and more and more, until there's absolutely no more aura or gold left to plunder, and then just move on to greener pastures to do it all over again!

(honestly, good timing, the king of terrorists is just insufferable, no?)

[사과나무 @treeapple33]

@jukan05 아지노모토도 공급 충분하다고 쇼티지 없다고 얘기했던거같은데 언젠가... 🤔

[사과나무 @treeapple33]

@jukan05 ISC도 충분히 포함 될 만한 것 같은데 말이죠..?🤔🤔

[사과나무 @treeapple33]

@jukan05 평생에 다시 오지않을 수 있는 쇼티지니... 제대로 뽕뽑아야 ㅋㅋㅋ

[사과나무 @treeapple33]

@jukan05 내용에 매우 동의합니다만, 그 LTA의 견고함이 확인되려면 다운사이클에 들어가봐야 아는게 아닐까요? 계약구조가 밝혀지더라도 시장에서 이를 쉽게 믿지는 않을 것도 같습니다 ㅎ

[사과나무 @treeapple33]

@jukan05 🥳🥳

[사과나무 @treeapple33]

@NathanYJLee 아직도 창구 수급보면서 모건이 사라고 리폿내놓고 팔았네 뭐네 하는 개미들이 10명 중 8-9명이라는 사실.. 그런 곳에 레버리지를 풀어놨으니 🫠

[사과나무 @treeapple33]

@jukan05 Capex는 모르겠고 계좌가 cut 되고 있긴합니다만 😂