Aaron

I still recall the skepticism when I first mentioned that $INTC was making solid progress with EMIB-T, and that several CSPs were actively considering it for their custom ASICs. There was quite a bit of pushback at the time, but it’s interesting to see how quietly that noise has faded as the supply chain reality sets in.

wow, congrats 🎉

@Aiden14544717 no comment:)

@Aaronwei3n They should learn from Aaron 🤣

JPM finally increase its $AMD and $GOOG shipment forecast...

@Aiden14544717 much more accurate comparing to 3 months ago... Agree with you on NVIDIA's.

@Aaronwei3n Accurate? TPU numbers is still a bit low, and Nvidia number looks overshoot to me

Chinese Model Distillation: Anthropic Evidence There are some people that claim that because Kimi K3 beats US models on some particular benchmarks, Kimi K3 didn't use distillation. Some even make jokes on when US models will be distilling from Chinese models. This is misconstrued. A smart student can cheat off a smarter student and then correct some of the answers from their own knowledge. Chinese researchers are no doubt amazingly smart and will likely make improvements on top with post training techniques. They aren't a dumb student cheating off a smart student. Chinese distillation attacks have gotten more advanced but there is un-refutable evidence published by Anthropic as recent as Feb 2026, that Chinese Labs were cheating: (1) anthropic.com/news/detecting… Anthropic does a good job on both technical and non-technical evidence. A funny one being: "We also observed tasks in which Claude was used to generate censorship-safe alternatives to politically sensitive queries like questions about dissidents, party leaders, or authoritarianism, likely in order to train DeepSeek’s own models to steer conversations away from censored topics" (1). A more serious evidence is "DeepSeek generated synchronized traffic across accounts. Identical patterns, shared payment methods, and coordinated timing" (1). Nowadays distillation techniques have gotten way more advanced with Chinese researchers leading in this area. In this recent June 2026 paper (2), Tencent researchers have done a survey of the on-policy distillation techniques. They claim it's for using older models to train newer models. Sure they might do that for some post training techniques but we all know what other models they use. (2) arxiv.org/pdf/2604.00626 As it relates to investing: 1. Yes Chinese labs skip buying GPUs to get very close to frontier models. 2. The Chinese model itself is free but the amount of GPUs you need to run inference (query the model) is very close to the same as US frontier models (3). Thus it doesn't matter if you use Chinese or American models, you need to buy the same amount of GPUs to run it. 3. US robots are very unlikely to run Chinese models. Sure the Chinese model runs on American compute on American soil but there are backdoors and biases that can be embedded in the model (4). The US government should not allow a huge fleet robots deployed in America like that.

@computeberg then I will miss everyone...😭

@Aaronwei3n what about -100%?

The stock drops 10% the first time, leaving you with 90%. It drops another 10%, leaving you with 81%—a loss of only 9% of your original capital. It drops a third time by 10%, leaving you with 72.9%—this time you’ve lost only another 8.1% of your original capital. Therefore, the more you’re down, the less you lose. 股票第一次跌 10%剩 90% 第二次再跌 10%剩 81% 只少了原本的 9% 第三次再跌 10%剩 72.9% 也只少了原本的 8.1% 因此當你虧得愈多 你就虧得愈少

lol, I am dead. Even a cancer support group has more hope than $MU.

@musk_eteers I am just trolling lol

@Aaronwei3n Yes but on the way up now, if it increases 10% then I gain even less, only 7.29% of original investment, and with 3 upward jumps of 10% I only go back to 97% So it works both ways, multiple 10% downs are difficult and lets just acknowledge that.

@Aiden14544717 嗚嗚

@Aaronwei3n 希望暴风雨后看得见彩虹😇

@truetrue2355 yes, of course, did Memory or Semi industry change comparing to 4 weeks ago? I believe the answer is no.

@Aaronwei3n The volatility is just a math puzzle; the fundamentals remain intact.

@TradeMeOrFold 😭😭😭

@Aaronwei3n Girl math

@Aiden14544717 對，上去兇 下來的也很兇（而且從上面跌更痛

@Aaronwei3n 这次的去杠杆是全球的啊

@WillLee88400940 謝謝台北李先生

@EternalEila GM

@Aaronwei3n Good morning!

Buyside overshoots every time. Still an impressive quarter from $TSM, especially under the circumstances of the 2nm GM dilution.

Morgan Stanley: CXL Market "We now raise our CXL market-size forecast in light of industry adoption progress and stronger demand from the elevated memory shortage. The updated estimate is US$2.1bn for CXL MXC chips and US$1.9bn for CXL switch chips in 2030, compared with previous estimates of US$990mn and US$664mn, respectively." $PENG $MRVL $ALAB

@truetrue2355 GM Hello Kitty

@ExtraMeatPizza 🤧

@Aaronwei3n Tell me about it, and my NVDA… it pains watching other small stupid shits go up a lot this year.

My heart is broken 💔

Does $TSM only deserve 17x?