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: 【量大強漲股整理】 GTC 驚天解密：代理式 AI 時代來臨！，四大題材股報你知!
TAIEX surged 605 points to a record 45,338 on NT$1.48T turnover as Jensen Huang's GTC Taipei keynote and pre-COMPUTEX momentum fueled broad AI buying; foreign investors net bought NT$36.8B. MediaTek (2454) jumped 5.68% to a new all-time high after announcing co-developed RTX Spark processors for Windows 11 AI PCs with NVIDIA, with a US broker reiterating Buy at NT$5,000 TP citing an early-stage AI ASIC up-cycle through 2028. AI server names Quanta (2382), Wistron (3231) and Inventec (2356) hit limit-up while Wiwynn (6669), Hon Hai (2317) and TSMC (2330) extended gains.
Why it matters: Names a concrete catalyst — MediaTek's NVIDIA RTX Spark AI PC chip launch with a NT$5,000 broker target and an analyst-flagged multi-year AI ASIC up-cycle — alongside limit-up moves across major Taiwan AI server suppliers.
Original: 〈輝達GTC〉Vera Rubin全面進入量產 黃仁勳大讚台灣150家供應商
Nvidia CEO Jensen Huang announced at GTC that the Vera Rubin platform — its largest-ever POD-class AI system, delivering up to 10x the agentic throughput of Grace Blackwell — is in full mass production, with 150+ Taiwan partners across 350 factories ramping output. Named system/ODM beneficiaries include Foxconn (2317), Compal (2324), Inventec (2356), Gigabyte (2376), Asus (2357), Pegatron (4938), Wistron (3231) and Wiwynn (6669), plus Spectrum-X CPO Ethernet switches now also entering production.
Why it matters: Nvidia explicitly named multiple in-universe Taiwan ODMs and server makers as Vera Rubin mass-production partners, a concrete revenue catalyst for the AI server supply chain.
Original: 台積電導入輝達 CUDA-X 服務與 AI 技術，加速新世代 AI 晶片開發製造
TSMC is deeply integrating NVIDIA's CUDA-X libraries and AI tools (cuLitho, cuEST, cuML, Metropolis, TAO, Omniverse) across lithography, transistor simulation, process control, defect inspection, and a digital-twin 'FabTwin' fab. NVIDIA cites 20–50% cost/cycle-time gains in computational lithography and ~50x speedup in materials simulation, deepening the TSMC–NVIDIA stack lock-in and reinforcing TSMC's leading-edge manufacturing moat ahead of next-gen AI chip ramps.
Why it matters: Strategic tech partnership deepening NVIDIA-TSMC integration with concrete productivity metrics, but not a discrete capex/contract event that immediately repricies the stock.
Original: [GTC 타이베이 2026] 엔비디아, 베라 루빈 시대 개막…젠슨 황 "본격 양산 돌입"
At GTC Taipei 2026, NVIDIA CEO Jensen Huang announced that the next-generation Vera Rubin platform has entered full-scale mass production, signaling the start of a new AI accelerator cycle. The launch reinforces NVIDIA's roadmap cadence and pulls in TSMC, HBM suppliers, and advanced packaging partners as the primary beneficiaries of the volume ramp.
Why it matters: Mass production kickoff of NVIDIA's next-gen Vera Rubin platform is a concrete product milestone with direct volume implications for TSMC, HBM suppliers, and advanced packaging partners.
Original: [컴퓨텍스 2026] 에이전틱 AI 시대, 다시 무대 중앙에 선 CPU
Computex 2026 spotlights the CPU's renewed strategic role as agentic AI workloads demand stronger orchestration and on-device inference, shifting attention back to general-purpose compute alongside accelerators. Intel, AMD, Qualcomm and Arm are positioned as primary beneficiaries as new CPU architectures target agentic AI PCs, edge inference, and AI server host processors.
Why it matters: Sector-wide theme highlighting CPU's resurgence in the agentic AI era, broadly impacting major CPU vendors without a single discrete catalyst.
Open source articleOriginal: 엔비디아, 애플 제치고 TSMC 최대 고객 등극…AI 발주 폭주
Korean media reports that Nvidia has surpassed Apple to become TSMC's single largest customer, reflecting the scale of AI accelerator demand routed through TSMC's advanced nodes and CoWoS packaging. The shift underscores TSMC's leverage in AI infrastructure capex and reinforces Nvidia's dominance of the AI silicon cycle.
Why it matters: Confirms ongoing AI demand concentration at TSMC/Nvidia rather than a discrete new event, but the customer-ranking shift is a meaningful sector signal.
Open source articleOriginal: 엔비디아 GTC 타이베이: 젠슨 황, 차세대 'Vera Rubin' GPU·'Vera' CPU 공개
At GTC Taipei, Nvidia CEO Jensen Huang introduced the new Vera Rubin GPU platform and Vera CPU as the core engines of the next compute revolution. The announcement reinforces Nvidia's AI infrastructure roadmap and signals continued pull-through demand for TSMC advanced nodes/CoWoS, HBM suppliers, and Taiwanese ODM/server assemblers in the Nvidia supply chain.
Why it matters: Nvidia's keynote unveiling of next-gen Vera Rubin GPU and Vera CPU is a direct product-roadmap event that anchors demand for TSMC, HBM, and Taiwanese AI server supply chain.
Original: GTC 타이베이: 엔비디아, Vera Rubin·Vera CPU 로드맵 차질 없음 재확인
At GTC Taipei, Nvidia confirmed its next-generation Vera Rubin GPU platform and Vera CPU remain on schedule, reinforcing the AI accelerator roadmap. The reaffirmation supports continued capex visibility for HBM, advanced packaging, and foundry partners through 2026-2027.
Why it matters: Direct roadmap confirmation from Nvidia on next-gen GPU/CPU products with material read-through to HBM suppliers, TSMC, and CoWoS supply chain.
Original: 〈台股盤後〉GTC大會RTX Spark亮相點爆AI PC股 飆漲604點收45337點新高
The TAIEX closed at a record 45,337.91 (+604.97 pts, +1.4%) on T$1.4T turnover after Jensen Huang unveiled the RTX Spark laptop at GTC, reigniting AI PC demand ahead of COMPUTEX 2026. MediaTek (2454) jumped ~5% to NT$4,555, Wistron (3231), Asus, Compal and Acer hit limit-up, while passive component names and Formosa Plastics group rallied on AI server pull-in and group restructuring.
Why it matters: Broad market wrap driven by an AI PC product launch and COMPUTEX anticipation — moves sector sentiment and named supply-chain names but is not a single stock-specific catalyst.
Original: 〈輝達GTC〉台達電張訓海：固態變壓器最慢明年迎來爆發
Delta Electronics (2308) President Chang Hsun-hai said at NVIDIA GTC Taipei 2026 that solid-state transformers (SST), developed over the past 5-10 years, will see a market breakout by 2027 at the latest, complementing the current 800V power system for AI data centers. SST's controllable architecture replaces multiple conversion stages, integrating and displacing portions of Delta's existing power product lines, while AI-assisted 'Smart Design' is expected to cut R&D cycles and costs by roughly 50%.
Why it matters: Roadmap/technology commentary from Delta on SST timing and AI-driven R&D efficiency — directionally positive for the AI power supply chain but lacks specific orders, capex, or earnings guidance.
Kioxia
285A
¥67,100
-12.86%