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: Rapidus 與英國半導體中心簽署合作備忘錄,強化雙邊半導體領域策略合作
Japan's Rapidus signed an MoU with the UK Semiconductor Centre (UKSC) during PM Takaichi's June 14 London visit, formalizing UK-Japan chip collaboration under the UK's AI Hardware programme. Rapidus reiterated its 2nm logic mass-production target for 2027 in Japan, with trial production already underway — a competitive signal versus TSMC and Samsung Foundry's 2nm roadmaps.
Why it matters: MoU is a diplomatic/strategic signal rather than a contract or capex commitment, but Rapidus's 2nm timeline directly competes with TSMC and Samsung Foundry, making it relevant supply-chain context.
Original: AI 引爆「記憶體超級循環」!外資看好美光居核心地位,2027 年獲利迎暴衝式成長
Aletheia Capital forecasts DRAM ASPs to rise ~30% QoQ in C3Q26 and another 10-15% in Q4, with HBM ASP more than doubling in 2027 vs 2026. Memory's share of AI hardware system value is seen rising from mid-40% in 2025 to 70%+ by 2027 (near 90% in high-memory configs), driving Micron EPS up 8.5x in CY27E and $350-400B cumulative FCF in FY26-28E — bullish read-across for SK Hynix and Samsung as HBM/DRAM peers.
Why it matters: Third-party sell-side forecast on memory ASP trajectory and a non-tracked US name (Micron); no named contract, capex, or earnings event for tracked tickers, but direct sector read-across to KR HBM/DRAM duopoly.
Open source articleOriginal: 傳高通洽談收購 Tenstorrent,交易金額落在 80 億至 100 億美元
Reuters, citing The Information, reports Qualcomm is in talks to acquire Jim Keller-led AI accelerator startup Tenstorrent for $8-10B, sending Qualcomm shares down ~1% after-hours. The deal, still subject to negotiation, would mark one of Qualcomm's largest AI-related M&A moves and accelerate its push into AI data center and inference markets against Nvidia, AMD and Intel.
Why it matters: US-centric M&A story with no direct Korean or Taiwanese chipmaker named; impact on tracked TW/KR names is indirect via AI accelerator competitive landscape.
Open source articleOriginal: 對決輝達!AMD 上市 Ryzen AI Halo 開發者平台,搶攻養龍蝦市場龐大商機
AMD started selling its Ryzen AI Halo developer mini-PC at $3,999, undercutting Nvidia's $4,679 DGX Spark while claiming 12-14% higher LLM token throughput on 30B-122B models and supporting up to 200B-parameter local inference via a Strix Halo SoC with 128GB LPDDR5X. A PRO 495 refresh with 192GB memory and 300B+ model support is slated for Q3 2026, escalating AMD's push into the on-device AI developer workstation segment against Nvidia and Apple.
Why it matters: Product launch with roadmap implications for on-device AI workstations and LPDDR5X memory demand, but no direct capex, contract, or named supply-chain beneficiary in the tracked TW/KR universe.
Open source articleOriginal: 〈台股開盤〉台股早盤大漲300點後翻黑 聯發科、鴻海撐盤 國巨一度觸及千元
TSE opened at 45,500 and rallied over 300 points on US-Iran ceasefire optimism before turning red as TSMC (2330) fell ~0.5% to NT$2,360. MediaTek (2454) rose ~1% to NT$4,510 and Hon Hai (2317) gained ~0.5%, while passives surged — Yageo (2327) jumped ~6% and briefly touched NT$1,000. Early-session turnover estimated at NT$1.5T.
Why it matters: Broad market open recap with sector rotation (passives strong, TSMC weak) — informative for TW semi exposure but no single stock-moving catalyst.
Original: AMD 宣布併 AI 記憶體優化商 MEXT、股價飆史高
AMD said it acquired MEXT, an AI memory-tiering specialist whose Predictive Memory Engine offloads cold data from costly DRAM to cheaper NAND flash without performance loss, aimed at easing rising memory costs for datacenter customers. AMD shares jumped 6.98% to a record close of $547.26 (up 155.54% YTD), pushing market cap to about $895B, within striking distance of $900B. The deal signals hyperscaler-facing efforts to soften HBM/DRAM cost inflation, a directional negative-to-neutral for memory makers' pricing leverage on AI workloads.
Why it matters: M&A is confirmed and AMD hit a record, but the deal is a small software tuck-in with undisclosed terms; the read-through to tracked Korean memory names (Samsung/SK hynix) and Taiwanese NAND/controller plays is directional rather than an immediate stock-moving catalyst.
Original: 市場估台積電先進製程不排除漲價,使三星受惠
Korean media report (via DDaily) flags that TSMC's 2nm node pricing is set to step up materially versus 3nm on EUV and advanced-packaging costs, pushing Nvidia, Apple and Qualcomm to diversify. Samsung Foundry, leveraging GAA at 2nm/3nm and more flexible pricing, is positioned to win incremental orders in automotive, robotics and edge-AI chips even as flagship AI silicon stays at TSMC.
Why it matters: Sector roadmap and pricing-dynamics story citing analyst/market commentary rather than a confirmed contract win or capex decision, but with clear directional read-through for Samsung Foundry vs TSMC competitive positioning.
Open source articleOriginal: 三星坦承 2027 年晶圓代工仍難獲利,先前發放巨額獎金是原因之一
Samsung DS chief Han Jin-man told staff the foundry business will still lose money in 2027, with breakeven now pushed to 2028, citing weak 2nm yields, mobile-skewed customer mix, low-margin orders, and a performance bonus pool equal to 10.5% of annual operating profit. Samsung remains a backup rather than first-choice TSMC alternative until 2nm yield clears the 70% threshold, while Qualcomm SoC discounts further pressure the Exynos 2700 launch.
Why it matters: Direct guidance from Samsung's DS chief that foundry breakeven slips to 2028 is a material stock-moving statement, with read-through to TSMC's competitive moat.
Open source articleOriginal: 〈台股盤前要聞〉台股2天狂漲2247點直攻4萬6、台積電面板級封裝PLP晶片明年量產
Korean media reports TSMC (2330) is building a Panel-Level Packaging (PLP) mass-production system, with a pilot line running this year and volume production targeted for next year, having already secured a major global AI chip customer — putting it in direct competition with Samsung (005930) for advanced packaging leadership. Separately, Hon Hai (2317) announced a strategic partnership with Schneider Electric for next-gen AI data centers, while TAIEX surged 1,228 points (+2.78%) to 45,397 on US-Iran deal news, with foreign investors net buying NT$46.5B (~$1.5B).
Why it matters: TSMC PLP mass-production roadmap with named AI customer directly challenges Samsung's advanced packaging position — a clear stock-moving competitive event for both names.
Open source articleOriginal: 韓媒:台積電面板級封裝PLP晶片明年量產 與三星競爭領先地位
Korea's etnews reports TSMC is accelerating panel-level packaging (PLP) buildout, with a pilot line running this year and mass production targeted as early as next year to challenge Samsung, which has led PLP since acquiring SEMCO's PLP unit in 2019. A 600x600mm panel yields 5-6x the chip output of a 300mm wafer, making PLP critical for large, complex AI dies; TSMC has reportedly already locked in a key global AI chip customer and is in talks with MCE suppliers on equipment investment.
Why it matters: Packaging roadmap/competitive story without a dated contract, capex figure, or named customer—directionally important for TSMC, Samsung Foundry, and OSAT players but not an immediate stock-moving event.
Open source articleJul 10, 2026 close · day-over-day
WOLF
$35
-5.26%