This morning I wrote why I stayed on X as a 30-year semiconductor engineer. I do not float random tickers and numbers as I walk by. I invest where I work. I put real money only in the names I can defend with people still on the floor. Then I come back to those same names and share what I actually know. That is the account. Since I came back to X in January I have posted the same short list — tickers, company names, and Korean. $NVDA $DELL Memory: $DRAM (listed Apr 2), SK hynix / $SKHY, $MU, SanDisk $SNDK $CBRS (IPO May 14) $AMD $AVGO Tape since Jan 1, except the two that listed this year (from listing day): $NVDA +17% YTD $DELL +350% YTD, about +11% today $DRAM +117% since Apr 2 $MU +240% YTD $AMD +140% YTD $AVGO +5% YTD $CBRS about flat from the $185 IPO, deep red from the day-one print $SKHY ADR listed in July — my book is still green $SNDK one of the strongest names in my own memory sleeve $NVDA Still the center of the board. Everything else I own is a satellite around this stack. My own cost basis is about 16x. Almost all of the position is gain, not principal. I do not need a taxable sale, and I have not found a better home for that capital. Custom silicon is cost control. It is not a full NVIDIA replacement. $DELL This is not a PC ticker. I used to own $SMCI as the one-top NVIDIA AI server name. Then the label-swap mess. I sit near people who have known that CEO and VP for decades. I heard the dawn FBI story as it landed. A stock can still run. I sold the whole position. Too many other choices to stay where I no longer trust the people. I moved it into Dell. Pure U.S. name. Trump has said "go buy a Dell" more than once, and it shows up in his disclosed book. I am not here for XPS. I am here for Dell as an AI server company — $95B backlog, Vera Rubin racks with CoreWeave. Today is another +11% session. Thesis has not changed. Memory I stay long the bottleneck. $DRAM I bought on listing day, April 2. As a U.S. investor I did not have a clean single ticker for Samsung + SK hynix + Micron. The basket let me put real weight on it. Over +100% since listing. I still tell Korean investors not to copy that ticker. Buy the Seoul names at home and keep $MU / $SNDK for the rest. Then IBKR omnibus for SK hynix ordinary shares, $SKHY ADR, $MU and SanDisk as singles. HBM yield, 1c mix, CPO, the new San Jose building — that is why SK hynix gets so many of my posts. Micron is the U.S. third leg. SanDisk fills the NAND hole the ETF underweights. Memory stays strong for a while. Nvidia's own call keeps repeating the same constraint. $CBRS This is the one that still hurts in the book. I was in on the idea before the IPO. The deal was too hot. I got a sliver, then bought more in the open market. That average is now my largest loss. I still praise the architecture as an engineer. Harvest, not a perfect wafer. Defect-tolerant tiles. Complementary to GPUs, not a replacement. That is why I keep writing it next to NVIDIA and memory. I visit Cerebras HQ next week, including the server room. I want to see the idea closer than a slide. $AMD Almost the only company that owns GPU, CPU, and ASIC in one house. Second in GPUs to NVIDIA. Second in CPUs to Intel. That full stack is the point. The Xilinx FPGA franchise is still the prototyping top for chips in front of those GPUs and ASICs. $AVGO Cable-modem days, then WiFi and Bluetooth. VMware made it a software company too. This cycle the payout is the ASIC / SoC design-partner seat — Jalapeño and the custom-silicon wave. Everyday life stays. These names stay. Not financial advice. I own what I write. I write what I work on.

This is pretty good. With Xfinity, when the promo ends the bill jumps unless you cancel and re-sign. I used to get it fixed by pestering support. Saw this post, remembered I’d written Muse off as useless and left it sitting there. Should actually try it. Usually the product isn’t the problem. The user is.

미국 계정의 경우 프리미엄 플러스는 예금 이자 6프로, 그냥 프리미엄은 4프로라서 플러스가 플러스됩니다 (물론 4프로에서 6프로로 direct drposit 설정하면 부스트업 할 방법이 있긴 합니다만)

Happy Chuseok long weekend 😎

First payout under X Original Content Rewards. Fewer posts and impressions than my last Revenue Sharing window. Higher payout for original content. Deposited to X Money.

The biggest beneficiaries of AI agents will be software businesses that are chronically short of skilled hands — and EDA is one of the clearest cases. The EDA market is still dominated by three firms that together control most of the industry: - Synopsys (SNPS): the leader in digital design flows (synthesis, implementation, signoff). Fusion Compiler, VCS, and PrimeTime are industry standards. - Cadence (CDNS): strongest in analog/custom design (Virtuoso) and digital implementation (Innovus). - Siemens EDA / Siemens AG (SIE.DE, SIEGY): owns the de facto standard for physical verification with Calibre, plus DFT. Synopsys (SNPS) stands out right now. After the Ansys deal it combined chip design with multiphysics, and it has been the most explicit about an L1–L5 agentic AI roadmap. Recent work with TSMC also pushes agentic workflows into analog, digital, and multi-die design. I wrote this up after attending Converge 2026 (formerly SNUG) in March. Resharing that piece here: 👇

정형외과 의사로서 추석에 부모님 댁에 가시면 꼭 하나 해드리라고 추천 드리고 싶은게 있습니다. "허리가 아프다", "무릎이 아프다" 하신다고 병원부터 모시고 가는 것도 중요하지만, 그보다 먼저 집 안을 한번 둘러보세요. 높은 문턱, 미끄러운 발매트, 어두운 복도, 바닥에 늘어진 전선, 손잡이 없는 욕실. 젊을 때는 별것 아닌 것들이 나이가 들면 (특히 거동이 좀 불편해지기 시작했다면) 낙상을 만드는 함정이 됩니다. 그리고 노년층의 낙상은 고관절 골절이나 척추 골절로 이어져 남은 삶 자체를 크게 바꿀 수 있습니다. 이번 추석에는 부모님 댁을 한 바퀴 돌면서 "어디 넘어지실 만한 데 없나?" 한번 찾아보세요. 비싼 건강식품보다 문턱 하나 없애드리고, 욕실 손잡이 하나 달아드리는 게 더 좋은 건강 선물일 수도 있습니다.

I have made the following paid articles free. I believe basic knowledge of optics, especially CPO, is something everyone should understand, so I decided to make all of my previously paid articles on this topic available for free. I can confidently say that if you read all of these articles, you will be able to build your own conviction on CPO. Please stop by my Substack, subscribe, and enjoy the articles below. The War of Light Has Begun: damnang2.substack.com/p/the-… The War of Light, Part2 - Who's at the Table, Who's Not, and Who's Hiding: damnang2.substack.com/p/the-… Why CPO Is Becoming inevitable? [CPO Special Part1]: damnang2.substack.com/p/why-… CPO, Fully Dissected [CPO Special Part2]: damnang2.substack.com/p/cpo-… The Illusion of CPO [CPO Special Final]: damnang2.substack.com/p/the-…

Four months later, that stack is a $1T company. May: Lisa Su said Venice was ramping on TSMC 2nm, and that $AMD’s edge was owning CPU, GPU, and ASIC in-house. Inference had put CPUs back in the middle of the room. (Below post 👇) Sep 21: AMD cleared $1T. Fourth U.S. chip in the club after Nvidia, Broadcom, Micron. Intel still sits near $640B. The market did not pay a trillion for a #2 GPU story. It paid for a company that can sell EPYC, Instinct, and a rack around both. Q2 data center $6.7B, more than double. More than half the company. Same thesis. Bigger number.

Is a ₩1.68 million Hermès poop-bag pouch excessive spending on a pet — or do they also have a right to share their owner’s wealth? Hermès treats the dog as “the rider’s second-best friend” and makes collars, leashes, walking pouches, and leather beds with the same saddlery craft as its bags. Examples: the Poopy pouch, the On-the-go dog pouch, and the Patapouf bed (large, $6,150). hermes.com/us/en/category/ho… Some say “that costs more than my dog.” Others say “this was never made for ordinary buyers.” The real question is simpler: affection, display — or both?

All green today except Tesla. The rest of the book is still doing the work.

Thanks for shouting me out with the great guys here. 🙏

World’s Top 15 Companies by Market Cap (as of Sep 24, 2026 close) U.S. names are Thursday’s close. Samsung is Seoul’s last session before Chuseok (Sep 23). NVIDIA remains first at $5.41T. Versus the Sep 22 board: • Meta passed SpaceX and took #7 at $1.96T • SpaceX slipped to #8, also $1.96T — a photo finish • NVIDIA faded to $5.41T • Samsung held #12 at $1.37T after a 3.3% Seoul close at 285,500 won • Five chip names still in the top 15: NVIDIA, TSMC, Broadcom, Samsung, Micron AMD is still outside the top 15, but it held $1T at #16. SK hynix remains next in line at ~$0.97T. Quiet tape after Monday’s spike. Meta was the rank change.

Thanks for shouting me out to the great guys here; it is also good to know you all to exchange tons of valuable things on X.

To someone outside the industry, "taping out" might sound confusing — but it's actually an old term from the days when we literally sent magnetic tape reels with the final chip design data. Back in 2000, right after I joined a semidonductor company, the president called me in and handed me a big reel-to-reel tape (the kind that looks like a video tape) saying, “This is the final tape for our previous generation of MIDI audio chip. Keep it safe.” That was the last physical tape I ever saw. Now we just transfer a digital file, yet we still proudly say “tape-out.” 😂 Congrats to the Tesla AI team on AI5 tape-out! After couple of decades the technology has completely changed, but the name lives on — a true industry classic.

Alibaba T-Head just unveiled Zhenwu V900. Company claims: - 3x performance vs. M890 216GB on-package memory - 1.2TB/s chip-to-chip bandwidth - Native FP8 / FP4 - Clusters up to 500,000 cards - Mass production target: 1Q 2027 - 20GW cloud DC capacity by 2032 Also teased Qwen models at 5–10 trillion parameters. What’s missing: process node, foundry, HBM generation/vendor, independent benchmarks. China’s AI stack is moving as chip + model + datacenter. The real bottlenecks remain memory, packaging, and power. #Alibaba #Zhenwu #AIChip #HBM #Semiconductors bloomberg.com/news/articles/…