Silicon Nexus

ASML publicly denied reports of exporting EUV lithography tools to China after US Commerce Secretary raised concerns, reaffirming that EUV remains off-limits under existing export controls. Chinese state media (VOA Chinese) framing highlights ongoing US pressure on the Dutch toolmaker and reinforces the domestic narrative that China must accelerate self-sufficient lithography development since SMIC and other foundries remain locked out of leading-edge EUV. The denial removes any near-term tail risk that China could legally access EUV, sustaining the structural advantage of TSMC and Samsung Foundry at sub-5nm nodes.

Why it matters: ASML EUV access is the single most important gating factor for China's leading-edge foundry capability; a formal denial reinforces TSMC/Samsung's sub-5nm moat and is directly material to SMIC-exposed supply chains.

Japan's METI unveiled a draft policy proposing rebuilding up to 5 aging nuclear reactors by the 2040s (and 11-14 by the 2050s) to secure stable power supply as AI-driven data center demand surges. The plan revives interest in nuclear-related equities (Japan Steel Works, Okano Valve, Sukawa Electric, Tokyo Energy & Systems, Taihei Dengyo), with the Takaichi administration's pro-nuclear stance also boosting next-gen reactor and SMR plays. Power infrastructure for AI/semis is positioned as a national strategic priority.

Why it matters: Japan's nuclear rebuild policy directly addresses AI/data-center power demand, a key bottleneck for the semi supply chain, though our tracked tickers are mostly indirect beneficiaries via power_infra theme rather than named in the article.

Amazon is reportedly exploring selling its in-house Trainium AI accelerators to third-party data center operators, expanding beyond AWS internal use. The move would position Trainium as a direct alternative to Nvidia's GPUs in the merchant AI silicon market, intensifying competition for hyperscaler and neocloud AI training workloads.

Why it matters: Hyperscaler custom-silicon commercialization is a sector-wide AI infra theme directly affecting Nvidia's merchant GPU TAM and TSMC/HBM supply chain, but the report is unconfirmed with no specific volume or timeline.

China is tightening oversight of indium exports, a critical minor metal used in compound semiconductors, display ITO coatings, and increasingly in AI chip packaging substrates. The move extends Beijing's pattern of weaponizing critical-material supply chains (gallium, germanium, graphite) and raises input-cost and supply-security risks for non-Chinese chipmakers and packaging houses as AI chip demand accelerates.

Why it matters: Sector-wide geopolitical supply-chain theme affecting compound semis and advanced packaging materials, but no immediate company-specific impact or quantified disruption disclosed.

A Korean industry outlook projects the global memory market expanding to roughly KRW 1,500 trillion, with AI server demand cited as the engine of a fresh semiconductor supercycle. The thesis reinforces structural tailwinds for HBM and high-density DRAM/NAND suppliers, primarily Samsung, SK Hynix, and Micron, but the piece is a market-sizing commentary rather than a discrete catalyst.

Why it matters: Sector-wide bullish outlook on memory/HBM driven by AI servers — supportive for major memory names but not a discrete near-term catalyst.

imec, ASML, and TSMC have jointly demonstrated complementary 2D-material transistors at a 50nm gate pitch on a 300mm wafer, a key milestone toward post-silicon CMOS scaling beyond the 1nm node. The breakthrough validates that EUV-compatible 2D channel devices can be built on production-grade tooling, though commercialization remains years away and does not alter near-term node roadmaps.

Why it matters: Long-horizon R&D milestone involving TSMC and ASML on next-generation transistor architecture; relevant to the equipment/foundry roadmap thesis but no near-term earnings or capex impact.

A Taiwanese NOR Flash memory supplier is rumored to have been designed into Nvidia's next-gen Vera Rubin AI platform, expanding NOR's role as boot/code storage in AI accelerator boards. Despite the bullish supply-chain news, foreign investors net-sold nearly 20,000 lots (~NT$4.1B) of the stock, signaling profit-taking or skepticism on the catalyst.

Why it matters: Vera Rubin supply-chain inclusion is a credible AI-infra signal for NOR Flash names, but the article does not confirm the specific supplier or order size.

AWS is shifting from internal-only Trainium/Inferentia deployment to selling its in-house AI accelerators externally, positioning to compete directly with NVIDIA and AMD in the merchant AI silicon market. The move signals hyperscaler vertical integration is deepening and could pressure NVIDIA's pricing power while creating new packaging/HBM demand routed through TSMC and Korean memory suppliers.

Why it matters: Hyperscaler ASIC commercialization is a sector-wide AI silicon competition theme affecting NVDA's moat and creating HBM/packaging demand routing, but no specific capex figure or near-term policy event is disclosed.

NVIDIA shares climbed to $310 on the back of sustained AI accelerator demand and record quarterly earnings, extending the 2026 rally. The move reinforces the bull thesis on AI compute spend and supports the broader HBM/foundry supply chain feeding NVDA's GPU ramp.

Why it matters: Price-action commentary referencing prior earnings rather than a new NVDA event or guidance change, but reinforces the AI compute demand theme relevant to HBM/foundry suppliers.