Global semi news — Korea, China, Taiwan, the US, and Japan. Government policy, export controls, capex moves, supply-chain shifts, and macro events. AI-classified and tagged with affected tickers. All headlines link back to the originating publisher.
Original: 傳美已批准向阿里巴巴／騰訊等 10 中企銷售 H200 晶片
The US Commerce Department has reportedly approved Nvidia H200 chip sales to about 10 Chinese firms including Alibaba, Tencent, ByteDance and JD.com, with Lenovo and Foxconn (2317) cleared as distributors and each approved buyer capped at 75,000 chips. Deals are stalled as Chinese firms withdraw under Beijing's guidance, but Nvidia CEO Jensen Huang joined Trump's China trip hoping the Xi summit breaks the impasse. Positive read-through for HBM suppliers tied to H200 if shipments resume.
Why it matters: Named policy beneficiary (Foxconn/2317 as approved H200 distributor) plus material read-through to Korean HBM suppliers if Chinese H200 shipments resume.
Original: 藉由光電整合，台積電揭示從 CPO 封裝到矽光子 COUPE 技術布局
TSMC detailed its silicon photonics roadmap at its annual tech symposium, with the first 200Gbps micro-ring modulator (MRM) using COUPE technology slated for mass production in 2026. The roadmap scales toward 400Gbps modulators and a 4Tbps/mm bandwidth density target by 2030, integrating EIC/PIC via SOIC packaging to replace copper in AI server interconnects. CPO/COUPE positioning reinforces TSMC's lock on advanced packaging for AI infrastructure, with ODM Quanta cited as a key ecosystem partner pushing optical-to-GPU integration.
Why it matters: Roadmap/technology disclosure with a concrete 2026 MRM production milestone, but no new capex, contract, or earnings catalyst — sector-level positioning rather than a stock-moving event.
Open source articleOriginal: 兩倍速晶圓廠擴建，台灣 2026 年新建四座晶圓廠及兩座先進封裝廠
TSMC plans to add four new wafer fabs and two advanced packaging plants in Taiwan in 2026 as part of an 18-fab global expansion (including five advanced packaging facilities), doubling its prior build pace to meet AI/HPC demand. Hsinchu Fab 20 and Kaohsiung Fab 22 are ramping 2nm, Taichung Fab 25 targets 2028 N2 production, while Arizona Fab 1 is set for 1.8x YoY growth in 2026 and Kumamoto output is guided to 20x the 2025 level.
Why it matters: Concrete TSMC capex roadmap with named fab counts, geographies and ramp multiples — direct read-through to 2330 and the broader semicap/packaging supply chain.
Original: 台積電發表最新製程與先進封裝藍圖，維持市場領先者地位
TSMC laid out its process and advanced packaging roadmap: N2 entered mass production in Q4 2025, N2P ramps 2H 2026, A16 with Super Power Rail is production-ready 2H 2026, A14 lands in 2028 and A13/A12 in 2029. On packaging, TSMC is producing 5.5x-reticle CoWoS at >98% yield, with 14x-reticle CoWoS integrating 20 HBM stacks targeted for 2028 and a >14x version with 24 HBM in 2029 — reinforcing TSMC's lead and underpinning HBM demand from Korean memory makers.
Why it matters: Official TSMC roadmap with concrete production timelines for N2/A16/A14 and CoWoS scaling to 24 HBM stacks — directly stock-moving for TSMC and tightly linked to HBM demand at Samsung/SK hynix.
Open source articleOriginal: 德州儀器調漲 PMIC、MOSFET，功率元件為何成 AI 關鍵？
Texas Instruments notified customers of 5-15% price hikes on PMICs, MOSFETs and industrial analog ICs effective July, driven by AI data center power buildout (400V/800V HVDC, BBU, high-power PSUs) as Nvidia's Rubin TDP hits 2300W and single AI racks approach 1MW. EU sanctions on China's Yangjie Tech accelerate de-Sinification of power semi supply chains, with Taiwan beneficiaries flagged including Niko-Sem (5340-equivalent), Sinopower, Anpec and Global Mixed-mode Tech on order migration.
Why it matters: Sector-level supply/demand story on power semis with broad AI data center read-through, but no named ticker in the tracked universe is a direct beneficiary (Taiwan names cited — Niko-Sem, Sinopower, Anpec, Global Mixed-mode — are not in the universe).
Open source articleOriginal: 不再單靠台積電？傳蘋果找英特爾代工 M7、A21 晶片
Apple has allegedly signed a foundry deal (late 2025) with Intel to produce the next-gen MacBook M7 chip on Intel 18A-P (mass production end-2027) and the iPhone A21 on Intel 14A (end-2028), diversifying away from TSMC amid AI-driven capacity crunch. If confirmed, this marginally reduces TSMC's leading-edge volume from its single largest customer, though TSMC remains dominant; Intel Foundry gains a marquee anchor customer alongside reported Google notebook collaboration.
Why it matters: Unconfirmed rumor on a multi-year roadmap shift (2027-2028 production) that could marginally affect TSMC's leading-edge mix but is not an immediate stock-moving event.
Open source articleOriginal: 聯電推出 14 奈米 eHV FinFET 平台，助力下一代 OLED 技術創新
UMC (2303) unveiled a 14nm embedded high-voltage FinFET process for OLED display driver ICs, validated at its Fab 12A, cutting power 40% and die area 35% versus its current 22nm eHV node. It marks the first FinFET introduction into the DDIC space and reinforces UMC's lead as the only foundry offering 22nm DDIC solutions, targeting premium and foldable OLED smartphones.
Why it matters: Process roadmap announcement strengthens UMC's DDIC franchise but is not an immediate earnings or contract catalyst.
Original: AI 熱、韓股年初迄今飆漲八成 避險基金續看多
Pertento Partners CIO Eduardo Marques calls Korea the most attractive value market, citing Samsung Electronics and SK Hynix as a 'golden duo' — SK Hynix trades at ~6x P/E with potential for a 10x rerating on sustained AI chip cash flows. Samsung and SK Hynix are up 128% and 188% YTD respectively; Korea's April exports topped $80B for a second straight month, up nearly 50% YoY on semiconductor strength. Interactive Brokers opened Korean equity and derivatives trading last week, broadening overseas retail access.
Why it matters: Sector-level bullish commentary from a named hedge fund PM with specific valuation thesis on SK Hynix and Samsung, plus supportive April export data — directional rather than a discrete stock-moving catalyst.
Open source articleOriginal: AI 有多熱，台積電：去年亞太晶圓用量堆疊逾 3 座 101 大樓
At its Taiwan Technology Symposium, TSMC said Asia-Pacific customers consumed over 2.1M 12-inch equivalent wafers last year, a stack ~1,600m tall — more than three Taipei 101 towers. The volume supported ~2,600 products in mass production and ~400 new product launches, underscoring AI-driven demand across cloud, edge, smartphones, and automotive.
Why it matters: Sector demand-color datapoint from TSMC management; directionally bullish for AI supply chain but no new capex, guidance, or contract figures.
Original: AI 驅動的半導體新紀元，台積電技術論壇揭示 1.5 兆美元產值藍圖
TSMC SVP Y.P. Chiang projected global semiconductor revenue to surpass $1T in 2026 and reach $1.5T by 2030, with AI/HPC taking 55% share. He highlighted 3DIC DRAM-on-logic stacking with memory partners, co-packaged optics (CPO) for high-speed links, and migration of automotive compute to 3nm/2nm — signaling sustained demand for advanced nodes, HBM, and advanced packaging across the TSMC ecosystem including Quanta.
Why it matters: Roadmap/market-outlook story from TSMC's tech forum with sector-wide implications for advanced nodes, HBM, and packaging — directional rather than a discrete stock-moving catalyst.
Open source articleKioxia
285A
¥67,100
-12.86%