Hugo

📊 semiconductor analyst

[Hugo @Hugo_Atn]

@pequityresearch Thanks for sharing the source! Yole Group's equipment team is insanely good 🔥

[Hugo @Hugo_Atn]

@IBM just announced the first sub 1nm node. 0.7nm, or 7Å.
Nanostack bonds two wafers to stack transistors vertically.
The cross section image is just wow!

[Hugo @Hugo_Atn]

OpenAI unveiled Jalapeño (Titan), its 1st custom inference AI ASIC designed with Broadcom.

[Hugo @Hugo_Atn]

Access restrictions are motivating China more than ever to develop their « EUV ». Thinking China will sit back and do nothing is utter hypocrisy. That’s the game. ASML and the Western world have slowed China down, but sooner or later the balance will return.

[spor @sporadica]

correct me if i'm wrong, but China getting their hands on an ASML euv machine is a WAAAAAYYYY bigger deal than like some Nvidia chips finding their way over there, fuck

[Hugo @Hugo_Atn]

If this is true, then it’s very big. Remember the Airbus A320 that mysteriously disappeared in the 2000s.

[Jukan @ICML @jukan05]

Holy shit, the U.S. government suspects that ASML exported EUV tools to China, or exported EUV components to China.

[Hugo @Hugo_Atn]

For fun I just asked @claudeai to develop a tool for processor die floorplan analysis, and the result is just crazy 🫪… I can’t imagine a funnier Friday evening 🤓

Btw, can you recognize the chip? 👀 Just an example bc I can’t share the amazing chips we’re analyzing rn 😅

[Hugo @Hugo_Atn]

Yole 2nm #GAAFET #Samsung, best TEM picture I saw on this.
source: Yole Group, Linkedin account

[Hugo @Hugo_Atn]

https://t.co/UxGoWcYYTi

[Hugo @Hugo_Atn]

Yole Group Intel 18A GAAFET and BSPDN PowerVia.
As you can see on the amazing die floorplan, with the metal present on both sides of the transistors, dense routing partially masks the underlying logic and memory structures. Link in the comment section

[Hugo @Hugo_Atn]

@yuuki_ans No it’s not that big, here you only see the package. You have to remove the heat spreader and you will see that the Grace CPU is smaller than the Hopper GPU die

[Hugo @Hugo_Atn]

(2/2)
Tesla vehicles from mid-2027 are expected to integrate the AI4+ chip, a slightly modified version by Samsung designed to double DRAM memory from 16GB to 32GB per SoC. The AI5 chip is still expected to be integrated into Tesla vehicles, but not immediately.

[Hugo @Hugo_Atn]

Quite a few updates on the Tesla/SpaceX/xAI roadmap:

(1/2)
1. The Terafab is expected to use Intel’s 14A node.

[Hugo @Hugo_Atn]

@braveben A deal for data centers in space maybe

[Hugo @Hugo_Atn]

@The_AI_Investor ASIC margin is less than that depending if the custom partner add a margin on top of the HBM or not… that’s a fundamental difference between Broadcom, Alchip, Marvell or MediaTek. If they don’t marge HBM but still buy it, then the company margin is impacted

[Hugo @Hugo_Atn]

@Teslaconomics Your AI4 picture corresponds to the infotainment board, not the ADAS…

[Hugo @Hugo_Atn]

@elonmusk @Tesla_AI Only Sk hynix for memory ?

[Hugo @Hugo_Atn]

@le20hfrancetele Non pas 1000 milliards l’année prochaine mais 1000 milliards sur la période 2025-2026-2027…

[Hugo @Hugo_Atn]

@RihardJarc Yes, and the impact won’t be only on Nvidia/AMD but on all the semiconductor supply chain that needs to adapt their investments of foundry and capacity based on the future demand… If you are interested, we just published an analysis on that: https://t.co/lF0kecNRtK

[Hugo @Hugo_Atn]

The success of #genAI and @OpenAI will inevitably hit a wall unless three conditions are met: Massive expansion of #power plants (1), massive CAPEX in #HBM memory (2), and an 8x surge in AI revenues within two years (3). (full analysis👇)
#Nvidia #CoWoS #TSMC #Wafer #GPU

[Hugo @Hugo_Atn]

https://t.co/4IZuh7LdZ7

[Hugo @Hugo_Atn]

The success of #genAI and @OpenAI will inevitably hit a wall unless three conditions are met: Massive expansion of #power plants (1), massive CAPEX in #HBM memory (2), and an 8x surge in AI revenues within two years (3). (full analysis👇)
#Nvidia #CoWoS #TSMC #Wafer #GPU