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: 폭스콘, 엔비디아 베라 루빈 구축비 공개…1GW당 1.5조 대만달러·연 전기료 410억
Foxconn (Hon Hai) disclosed that building out 1GW of Nvidia Vera Rubin AI infrastructure costs roughly NT$1.5 trillion, with annual electricity bills reaching NT$41 billion. The figures underscore the staggering capex and power demands of next-gen Nvidia AI systems, reinforcing the AI infra and power supply theme.
Why it matters: Sector-wide AI infra/power-demand datapoint disclosed by Foxconn on Nvidia's next-gen Vera Rubin platform; informs capex and power-infra theme without being a direct earnings event.
Original: 記憶體晶片短缺暫時無解,消費性電子產品將替 AI 產業買單
DRAM/NAND capacity at Samsung, SK Hynix and Micron is locked up by AI buyers, while US fabs (Micron Idaho mid-2027, NY 2030) and China's CXMT/YMTC remain blocked by national security concerns. Memory gross margins have hit record highs (Micron ~80%) but the Big 3 stay cautious on expansion after 2023 losses, leaving consumers — starting with Apple — to absorb price hikes.
Why it matters: Sector-wide supply/demand commentary reinforcing the AI-driven memory tightness thesis; bullish read-through for memory trio but no new contract, capex or policy catalyst.
Open source articleOriginal: 波若威:CPO產品H2量產出貨 光被動元件三大產品應用拚發光
Browave (3163-TW) said its 800G fiber shuffle kits and patch boxes qualified into Nvidia's CPO ecosystem will begin mass production in H2 2026, with small Q3 trial output, monthly capacity reaching the low thousands in Q4, and a target of 10,000+ units/month by Q1 2027. Growth is anchored on three passive-optical product lines — AI/DC high-density fiber patch boxes, 1.6T fiber kits for North America, and 800G/1.6T custom jumpers/MMC — with Shuffle Module shipments starting late Q3 as the key AI data-center entry product.
Why it matters: Concrete CPO supply-chain win and a quantified Q4 ramp for a non-tracked Taiwan optical name, with read-through to Nvidia AI infra demand but no tracked TW/KR ticker directly named.
Open source articleOriginal: 台積電7月法說成風向球!外資示警台股3大整理訊號「主動」選股價值浮現
Schroders warns of three correction signals in Taiwan equities — rapid margin-debt expansion, fading leadership in high-priced and large-cap names, and rotation into laggards and theme stocks — suggesting the market is shifting from trend-up to consolidation. TSMC's (2330) July earnings call is the focal point: consensus expects solid Q2 numbers, but unless full-year guidance is revised higher (à la Broadcom), profit-taking pressure could weigh on the stock. JPMorgan AM notes capital rotation last week into semis (+7.55%) and financials (+7.76%), while AI-adjacent comms/PC peripherals lagged.
Why it matters: Strategist commentary previewing TSMC's July earnings call and broader Taiwan market rotation — sector/market-data story rather than a confirmed stock-moving event.
Original: 〈熱門股〉南亞轉型電子材料題材噴火 5天漲4根停板飆得比南亞科還兇
Nanya Plastics (1303) surged 30.52% last week with 4 limit-ups in 5 days, outpacing affiliate Nanya Tech (2408), as AI electronic materials drove May revenue to a 47-month high and foreign investors net bought 68,000 lots over 5 sessions. The company plans to divest portions of its Nanya Tech and NYPCB (8046) stakes through year-end to fund ABF substrate upstream material capacity expansion, with electronic materials now over 50% of revenue under Formosa Plastics (1301) group's transformation plan.
Why it matters: Stock price commentary with concrete capacity expansion funding plan and revenue milestone, but core driver is already-known AI memory demand rather than a new catalyst.
Open source articleOriginal: 全球AI專用光收發模組規模今年增57% 智邦、明泰布局1.6T交換器動能強勁
TrendForce projects the global AI optical transceiver market will expand from $16.5B in 2025 to $26B in 2026 (+57% YoY), driven by hyperscaler AI datacenter buildout and the 800G-to-1.6T transition. Accton (2345) has begun mass production of 1.6T switches with H2 shipment ramp and is expanding capacity across Taiwan, Vietnam, US, Singapore and Malaysia; Alpha Networks (3380, not in tracked universe) targets Q2 sampling and year-end mass production of its 1.6T liquid-cooled switch.
Why it matters: Names a tracked ticker (2345) as a direct beneficiary with concrete production milestones (1.6T mass production, H2 shipment ramp, multi-country capacity expansion) tied to a quantified market sizing.
Original: AI與低軌衛星雙引擎引爆!網通與光通訊營收飆歷史新高
Accuselink (2345) May revenue jumped 56.58% YoY on 800G switch and AI accelerator shipments, with 1.6T switches ramping in H2. Sercomm (5388) posted ~70% YoY May growth and Wistron NeWeb (6285) +11.6% YTD on LEO satellite subscriber gains; only 2345 sits in the tracked universe.
Why it matters: Sector-wide May revenue update with bullish AI networking/LEO demand signals but no single capex, contract, or earnings catalyst, and most named beneficiaries sit outside the tracked universe.
Original: 〈觀察〉GPU及ASIC需求強勁 健策、奇鋐、雙鴻下半年營運看增
Jentech (3653), Auras (3017) and Asia Vital/Asia Vital Components' peer Auras flag stronger H2 than H1 as Nvidia GB-series shipments stay strong, Vera Rubin ramps and ASIC water-cooling volumes scale; Jentech expects several ASIC customers' monthly vapor-chamber shipments to exceed GPU from H2, while AMD MI455 two-piece vapor chambers ramp in Q4 at higher ASPs. Auras' chairman cites order visibility into 2028-2029 with Vietnam at ~45% of capacity heading toward a 50/50 split with China; Asia Vital Components peer Auras Tech (Shuanghong/3324) sees liquid-cooling reaching ~70% of revenue by year-end with ASIC liquid-cooling contributing up to 40%.
Why it matters: Sector/supply-chain update with management commentary and capacity color on AI thermal demand, but no single hard catalyst like a named contract, capex figure or guidance reset.
Open source articleOriginal: 폭스콘, 베라 루빈 데이터센터 GW당 구축비용 공개
Foxconn has reportedly disclosed a per-gigawatt cost benchmark for NVIDIA's Vera Rubin generation AI datacenters, providing a concrete capex anchor for next-gen AI infra buildouts. The figure implies sustained, large-scale spending on GPUs, HBM, networking and power infrastructure tied to Rubin deployments, reinforcing the AI capex cycle into 2026-2027.
Why it matters: Sector-wide AI infra capex signal centered on NVIDIA's Rubin platform and Foxconn's system integration, with read-through to HBM, networking and power names.
Original: 第三季財報優於預期,外資調升美光目標價達 1,300 美元
Wedbush reiterated Outperform on Micron and lifted its target from $550 to $1,300 (vs. last close $1,152), modeling Q3 DRAM/NAND ASP up 65% QoQ versus a prior 40% estimate after Q2 2026 memory prices rose by double- to triple-digit percentages. The analyst sees another ~20% price hike into Q4, with AI-driven demand keeping the market tight through at least 2027 — directly bullish read-through for Korean memory peers Samsung Electronics and SK hynix, whose contract prices ran above Micron's early-quarter levels.
Why it matters: Major sell-side target-price doubling with explicit Q3 DRAM/NAND ASP guidance raise reads directly into Korean memory peers' earnings and pricing outlook.
Open source articleJul 10, 2026 close · day-over-day
WOLF
$35
-5.26%