This latest piece from GSR is genuinely important. Why is Nvidia still courting China — almost clinging to it? GSR frames it this way: "For Nvidia, China is a monitoring window — a way to observe how massive AI infrastructure is actually evolving." Nvidia's revenue exposure to China has collapsed from roughly 26% historically to around 5% today. And yet the reason Nvidia cannot easily walk away from China is not about revenue. China's domestic AI chip competition is already fierce. Huawei Ascend, DeepSeek, Alibaba — China's homegrown AI ecosystem is no longer in the "chasing Western technology" phase. Under the pressure of sanctions, it has moved into genuine architectural experimentation and system-level optimization. In other words, China is no longer simply a sales market for Nvidia. It is the single most important live laboratory for observing how large-scale AI infrastructure evolves outside the Nvidia stack. GSR's argument, however, is that Nvidia's China organization is currently failing to execute on this strategic objective. According to the author, Nvidia's people on the ground in China are not listening to real customer needs or running field-driven sales. Instead, they are leaning on abstract, academic-style lectures — behaving less like a commercial organization and more like a "club of scientists." This is residual inertia from the era when Nvidia products sold themselves the moment they were available. If the core purpose of Nvidia's China strategy is intelligence gathering and on-the-ground learning, then the most important variable should be tight, two-way communication with customers. Yet Nvidia is failing to genuinely talk to the very market it is supposed to be observing. That is the fatal contradiction. The problem is already showing up in Nvidia's recent failures in China. The most representative case is the RTX 6000D. It was a hastily designed, downgraded GPU engineered to fit just under U.S. export control thresholds. According to GSR, Nvidia did not transparently communicate to its channel partners the critical missing specs — the absence of NVLink, the absence of HBM — that directly determined the product's competitiveness. Instead, the company defaulted to its old assumption: "if it has an Nvidia logo on it, it will sell." Volume was pushed through on that premise. The outcome was poor. Stripped of its competitive edge, the RTX 6000D was rejected by the market and ended up as massive dead inventory sitting on partners' books. The author's broader point is that Nvidia's China problem is no longer just an export control problem. It is a listening problem — a signal that the sales model built on Nvidia's overwhelming market dominance simply no longer works in this environment. An excellent read. $NVDA

Intel’s AI Comeback: From HPC Struggles to 3D Integration Leadership Intel, which had been struggling in the AI HPC domain, is now drawing significant attention alongside AMD as AI rapidly expands beyond training into inference and agentic workloads. This shift is increasing demand not only for massively parallel computation used in training, but also for sequential, composite operations required for inference ‘driving renewed CPU demand’. At the same time, the sharp rise in memory chip prices, combined with a market that is reacting more quickly to monetization through AI services than to further scaling of training, is accelerating this transition. In addition, growing demand for real-world deployment of Physical AI and Vertical AI is diversifying workloads and increasing the need for operational compute to coordinate and integrate memory and communication within workflows. In response, Intel is strengthening its CPU business while also developing and planning to unveil a more advanced 3D chip integration technology (HB3DM), which goes beyond TSMC’s 2.5D CoWoS solution ‘where supply bottlenecks and cost pressures are concentrated’ to deliver higher levels of computing integration. While it is still too early to judge its commercial success, this development can be seen as reflecting Intel’s intent to rapidly advance technology in the direction required by evolving AI computing paradigms and to regain leadership. The technology, expected to be presented at VLSI in June 2026, directly integrates eight DRAM dies with a single logic (xPU) die at the silicon level using hybrid bonding. Through approximately 13.7k TSVs, the memory chips and xPU are interconnected, enabling 5.3 TB/s of bandwidth . .. semihpc.com

ASML and labour unions have agreed on a social plan designed to minimise forced layoffs at the company. The semiconductor equipment maker said in late January that approximately 1,700 management roles would be cut, though a company spokesperson stated on Tuesday that the final number of job losses could be reduced by several hundred. nltimes.nl/2026/05/13/asml-r…

マイクロンCEOの中国訪問。 これは 「中国にハイエンドメモリを売り込みに行った」 というより、 2023年以降に悪化した中国での事業環境を、政治レベルで少しでも安定させる動き と見る方が自然だと思います。 関連銘柄： マイクロン、キオクシア、サンディスク、SKハイニックス、サムスン電子

변압기 없으면 태양광도 무용지물… 미국 전력망 ‘병목’ 비상 1/ 미국 전력용 변압기 공급난이 전력망 확장과 재생에너지 프로젝트의 새로운 병목 현상으로 부상했다. 2/ PV 매거진은 미국 고용량 전력 변압기 납기가 최대 4년까지 늘어났다고 보도했다. 재생에너지 발전소와 송전망, 데이터센터 수요가 동시에 폭증하면서 인프라 장비 확보가 어려워진 결과다. 3/ 변압기를 제때 구하지 못하면 발전소가 완공돼도 전력을 계통에 연결하지 못하는 상황이 발생할 수 있다. 발전사업자와 데이터센터 운영사 간 변압기 물량 경쟁이 치열해지면서 조달이 프로젝트 준공 시점을 결정짓는 핵심 변수가 됐다. 4/ 우드 맥켄지에 따르면 발전기 승압 변압기 수요는 2019~2025년 사이 274% 증가했고 변전소 변압기 수요도 116% 늘었다. AI 데이터센터와 전기차, 제조업 전환 등이 전력 수요를 밀어 올리며 변압기 가격은 최근 5년간 약 80% 상승했다. 5/ 공급 제약의 주요 원인은 방향성 전기강판과 구리 부족이다. 미국 내 생산 기반이 제한적이어서 수입 의존도가 높다. 6/ 히타치에너지는 10억 달러 이상 투자로 2028년 신규 공장을 가동할 계획이며 지멘스도 노스캐롤라이나 공장 투자를 확대했다. 7/ 재생에너지 개발사들은 변압기 확보가 금융 조달과 일정 관리의 핵심이 됐다고 지적했다. 앞으로 프로젝트 경쟁력은 변압기 생산 슬롯 선점 여부에 크게 좌우될 전망이다.

2D NAND不足の話、かなり重要。 無料公開情報で確認できる範囲でも、 ・サムスンは最後の2D NANDラインを1c DRAM向けに転換 ・キオクシアは2D NAND/旧世代BiCSを2028年出荷終了へ ・TrendForceは2026年のMLC NAND能力を前年比41.7%減と予測 ・マイクロンもCrucialを終了し、AI/企業向けへ集中 つまり、旧世代メモリは「安く作って供給し続ける」時代ではなくなりつつある。 ポイントは、2D NAND市場が巨大だからではない。 低採算品を切り、HBM・先端DRAM・3D NANDへキャパを振り向ける流れが鮮明になっていること。 これはメモリ株には基本的に追い風。 ただし、2D NAND価格急騰の直接恩恵は、キオクシアやマイクロンより、レガシー品を残す台湾勢に出やすい点は注意。 関連銘柄： キオクシア、マイクロン、サンディスク、SKハイニックス、サムスン電子、旺宏、華邦電

MSG maker Ajinomoto to raise prices by 30% as ABF substrate supply-demand gap extends to 2027 due to AI supercycle. wccftech.com/msg-maker-ajino…

This change reflects a deeper transformation in the semiconductor industry. AI workloads have pushed computing beyond the boundary of individual chips. Performance is now determined by the entire system: how chips are packaged, how memory is connected, how data moves between accelerators, how heat is removed, and how power is delivered across the rack. COMPUTEX has therefore become a place where the next generation of AI infrastructure is displayed, negotiated, and coordinated. tspasemiconductor.substack.c…

Samsung’s labor talks have collapsed, raising fears a strike could disrupt global memory chip supplies, media report, adding government mediators say they will spare no effort to broker an agreement. An 18-day strike could begin as early as next Thursday, May 21. Unions are demanding 15% of operating profits for bonuses and a removal of the cap. Samsung is offering 10% plus a 1-time special compensation packaging. $SSNLF $HXSCL $MU #Samsung #SKhynix #semiconductors #semiconductor koreaherald.com/article/1073…

Another company from my series on German hidden champions in photonics, without which the CPO ramp simply would not be possible. Another physical enabler. Employees: ~150. Visibility in the financial press: practically zero. Importance to the global transceiver industry: critical. Imagine having to place a laser diode the size of a grain of salt, roughly 300 micrometers wide, onto a submount with such precision that its optical output aligns perfectly with the core of a glass fiber. The core of that fiber is just nine micrometers wide, about one tenth the width of a human hair. Shift the diode sideways by a single micrometer and you lose roughly 30% of your optical power. Shift it by three micrometers and the entire module becomes scrap. And this is not a one off alignment inside a cleanroom lab. It has to be done thousands of times per month in mass production of optical transceivers. That is exactly what AMICRA was built for. And nobody in the world does it more precisely. Founded in 2001 and acquired in 2018 by ASMPT, the company builds what are arguably the world’s most precise die bonders for photonic packaging. The NANO platform places components with up to ±200 nanometer accuracy at 3 sigma, roughly one fiftieth the thickness of a human hair. No other commercially available die bonder achieves these numbers under real production conditions. Direct competitors, MRSI in Massachusetts, Palomar in California, and a handful of Japanese suppliers, typically operate in the 500 nanometer to 1 micrometer range based on public specifications, two to five times worse. In this business, every additional micrometer of misalignment disproportionately worsens coupling loss. This is not a marketing detail. It is the difference between a functioning module and scrap. These machines cost between €1.5M and €3M, weigh several tons, sit on granite based anti vibration platforms, and perform active optical alignment in real time. In a world where every 800G transceiver contains 60 to 100 such bonds, and every 1.6T module roughly doubles that number, this is no longer a niche business. It is foundational volume infrastructure beneath the CPO hype everyone talks about. In 2025, roughly 30 million 400G / 800G / 1.6T optical modules were produced globally, with the 800G and above segment expected to grow around 30% annually through 2030. Even classical pluggable transceivers, LPO modules, and the external laser sources used for CPO itself are assembled on NANO systems. Industry consensus estimates AMICRA’s market share in active alignment capable high precision bonders at roughly 40 to 50%, making it the undisputed global leader in its category. Unfortunately, AMICRA has been fully integrated into ASMPT since 2018. The systems are sold under ASMPT’s photonics division, and the photonics business still represents less than 10% of ASMPT’s HKD 14.5B revenue base. If you want exposure to AMICRA, you buy ASMPT, and the market has long since absorbed the TCB, HBM, and CPO optimism. AMICRA is technically dominant, industrially mature, indispensable to a structurally growing market, and yet today not really an investment case. Another example of a German hidden champion buried inside a larger international conglomerate. Personally, I have not seen a single person on X write about AMICRA so far. I only discovered the company myself during my research in late 2025.

AI frees you from having the quality of service depend on others' competency. It allows you to roll up your sleeves if you are not satisfied with the work. Thus, it terrifies others that people may discover how long they have been subjected to incompetence.

중국 시장조사업체 “2027년 AI 서버가 D램 비중 49%, 스마트폰 추월” 1/ 중국 시장조사업체 시그마인텔은 AI 서버 확대로 글로벌 D램 수요 구조가 크게 재편될 것이라고 전망했다. 2/ 스마트폰과 PC 중심이던 D램 수요가 AI 서버로 이동하면서 2027년 AI 서버용 D램 비중이 49%까지 상승해 스마트폰을 추월할 것으로 예측된다. 3/ AI 서버용 D램 비중은 2026년 40%에서 2027년 49%, 2028년 50~55%로 증가하고 스마트폰용은 2024년 43%에서 2027년 23%까지 줄어들 전망이다. 4/ 글로벌 AI 서버 출하량은 2026년 270만 대로 51.3% 증가하고 2028년에는 500만 대에 가까워질 것으로 보인다. 5/ 올해 AI 서버용 DDR 수요는 전년 대비 105%, HBM 수요는 110% 급증하며 고성장세를 이어갈 것으로 관측된다. 6/ AI 패러다임이 학습에서 추론으로 이동하면서 D램 수요 확대가 지속될 것이며 향후 3~5년간 전체 메모리 용량 수요도 꾸준히 증가할 전망이다.