Weekly Highlights

Samsung Taps TSMCs 3nm Process for Nvidia HBM Base Dies in Rare Tie-Up, Reshaping the HBM Value Chain

On October 19, multiple media reports said Samsung Electronics plans to use TSMCs 3nm process to manufacture base dies for Nvidias next-generation custom High Bandwidth Memory (NVHBM). Key details include:

• ●
  Three-supplier alignment: Nvidia is reportedly requiring Samsung Electronics, SK hynix and Micron to use TSMCs 3nm process for NVHBM base-die production to ensure performance consistency. This would mark Samsungs first cooperation with TSMC at the HBM manufacturing stage.
• ●
  Yield is critical: Industry estimates put TSMCs 3nm yield above 90%, versus roughly 60% for Samsung Foundrys 2nm SF2 process. For Nvidia, concentrating base-die production at TSMC offers a more stable manufacturing option.
• ●
  Technical significance: The base die is a core HBM component that carries logic and I/O circuitry and works with the stacked memory dies to form the complete HBM device. Moving the base die to an advanced node more tightly links the HBM value chain with logic foundry manufacturing.
• ●
  Industry impact: The move signals a structural shift in HBM manufacturing, with base dies moving from supplier-specific in-house approaches toward standardized advanced-node foundry production, allowing TSMC to penetrate deeper into the HBM value chain.
• ●
  Market reaction: Following the reports, TSMC (TSM.US) rose more than 3% in pre-market trading, reflecting a positive market response to the deeper integration of advanced-node foundry capacity with HBM.

This development sits at the intersection of HBM manufacturing and advanced-node foundry services, with the restructuring expected to cascade through the supply chain:

• ▲Upstream: TSMCs 3nm advanced-node foundry services for base dies, HBM memory-die production by Samsung Electronics / SK hynix / Micron, and advanced packaging materials and equipment.
• ▼Midstream: HBM integration and stacking/packaging using C2W/W2W bonding, plus AI accelerator-card integration. Standardizing the base-die process reshapes HBM yield and cost structures.
• ●
  Downstream: Nvidia GPUs, including the Vera Rubin platform, and AI servers. HBM supply stability directly affects the delivery pace of AI compute capacity.

The impact is expected to cascade through the value chain, with significant implications across multiple segments:

• ■Advanced-node foundry: Moving HBM base dies to advanced nodes opens a new HBM growth vector for TSMCs 3nm business and further strengthens its pricing power. Manufacturing integration across memory and logic is becoming a new norm in the AI chip supply chain.
• ■Memory manufacturing: HBM production at Samsung Electronics, SK hynix and Micron is becoming more closely tied to external foundry capacity. Memory suppliers lose some autonomy at the base-die stage, shifting the competitive focus toward memory-die technology and packaging integration.
• ■AI compute: Nvidia is driving supply-chain standardization to secure consistent HBM performance and stable supply, pushing the AI compute supply chain toward more platform-level integration.
• ■China industry perspective: Tying HBM base dies to advanced-node foundry capacity highlights the urgency of coordinated development across memory, logic and advanced packaging. The path for coordinated development of domestic HBM and advanced-node manufacturing is becoming clearer.
• ■Customs perspective: Import/export classification, control requirements and tariff arrangements for HBM wafers and finished products may evolve with changes in the supply-chain structure, making regulatory treatment of HBM-related items a new focus for customs compliance.

Update 1 | China Proposes Revising Catalogue of Technologies Prohibited and Restricted from Export, Bringing AI Chip Design Technologies Under Export Controls for the First Time

On October 16, Chinas Ministry of Commerce and Ministry of Science and Technology issued Announcement No. 48 of 2026, proposing revisions to the Catalogue of Technologies Prohibited and Restricted from Export and seeking public comments, while simultaneously introducing temporary pre-export registration for the relevant technologies. Key points include:

• ●
  Proposed additions: The draft would add 40 technologies across artificial intelligence, biomedicine and photovoltaics, including 23 AI-related technologies.
• ●
  Chip-design controls: AI processor chip design technologies involving 14nm and below process nodes, 2.5D/3D packaging or non-von Neumann architectures, as well as advanced 3D flash-memory and memory-chip design technologies, are proposed for inclusion among technologies restricted from export.
• ●
  Implementation: The announcement introduces temporary pre-export registration for the relevant technologies while soliciting public comments.
• ●
  Policy background: Based on the Foreign Trade Law of the Peoples Republic of China and the Regulations on the Administration of Import and Export of Technologies, the proposal aims to balance development and security while improving the technology export management framework.
• ●
  International developments: In parallel, Taiwan authorities are reportedly considering adding AI servers and advanced semiconductors to export-control lists for the Chinese mainland, indicating tightening technology controls in both directions.

The policy sits at the policy layer governing technology exports and intellectual-property controls and affects the entire chip-design value chain:

• ▲Upstream: Chip-design tools (EDA), IP licensing and design services. Export controls on technical know-how may affect cross-border collaboration models.
• ▼Midstream: AI chip design companies working with 14nm and below nodes, 2.5D/3D packaging or non-von Neumann architectures, as well as advanced 3D flash-memory and memory-chip design technologies.
• ●
  Downstream: Wafer foundries, packaging and testing, and system vendors, with export registration and licensing requirements affecting cross-border supply-chain arrangements.

• ■Chip-design know-how is being brought under export controls on a broader scale for the first time, subjecting cross-border transfers, licensing and collaboration involving advanced AI chip design and 3D memory design technologies to tighter scrutiny.
• ■Pre-export registration increases compliance costs and process lead times, requiring companies involved in technology exports to assess licensing routes and registration requirements in advance.
• ■Together with proposed restrictions by Taiwan authorities on AI servers and advanced semiconductors bound for the Chinese mainland, semiconductor technology controls are tightening in both directions, prompting a reassessment of cross-border supply-chain layouts.
• ■Customs perspective: Aligning regulatory treatment of technology exports, including software, drawings and know-how, with goods exports is becoming critical, while coordinated review of dual-use controls and technology export licensing is increasingly important.

Update 2 | CXMT 9M Revenue Rises More Than 700% YoY; Net Profit Turns Positive at RMB 131bn–132.5bn

From October 19 to 20, CXMT Corporation disclosed its earnings forecast for the first nine months of 2026, showing exceptionally strong growth. Key highlights include:

• ●
  Revenue: Revenue for the first nine months is expected to reach RMB 265bn–270bn, up more than 700% YoY.
• ●
  Net profit: Net profit attributable to shareholders is expected at RMB 131bn–132.5bn, turning profitable from a year-earlier loss and reaching a record high.
• ●
  Drivers: Improving industry conditions, high capacity utilization at the companys two fabs, and continued product-mix optimization.
• ●
  Shareholder meeting remarks: The company remains confident in the long-term outlook for the DRAM industry and continues to view the supply-demand dynamics of the memory supercycle favorably.
• ●
  Institutional attention: Following the earnings forecast, leading brokerage research teams and senior semiconductor industry experts showed heightened interest in the memory-sector outlook.

CXMT sits at the core of domestic DRAM manufacturing, with the earnings surge cascading through the supply chain:

• ▲Upstream: Domestic suppliers of memory manufacturing equipment, including etching, thin-film deposition and hybrid bonding tools, as well as silicon wafers, photoresists and other materials. Capacity expansion is driving equipment and materials orders.
• ▼Midstream: CXMTs DRAM wafer manufacturing, including G5 platform volume production and 4F² architecture development, positioning it as the worlds fourth-largest DRAM supplier.
• ●
  Downstream: AI servers, data centers, smartphones and PCs, with DDR5/LPDDR5X demand continuing to ramp.

• ■Revenue growth of more than 700% and net profit above RMB 131bn mark a new stage of profitability at scale for Chinas leading domestic memory supplier, as the benefits of the AI memory supercycle are fully reflected in earnings.
• ■High capacity utilization and product-mix optimization, including a rising share of DDR5/LPDDR5X, are accelerating domestic substitution in high-end memory and supporting continued gains in global DRAM market share.
• ■Volume production of the G5 platform and breakthroughs in 4F² architecture support long-term technology competitiveness, moving domestic memory from scale expansion toward technology leadership.
• ■Customs perspective: As both volumes and prices rise for high-end memory chips, HS classification, origin determination and port-inspection turnaround times for DDR5/LPDDR5X are becoming key variables in the memory supply chain.

• 1

  Nvidia Partners with Six Upstream Suppliers to Expand Capacity; Samsung Electro-Mechanics Joins Package-Substrate Supply Chain

  Event: On October 22, Nvidia announced cooperation with six upstream suppliers to expand AI infrastructure capacity, adding capacity equivalent to roughly 2x current levels, with the new capacity expected to begin coming online from Q1 2028. Samsung Electro-Mechanics will join Nvidias substrate supply chain and begin shipments immediately. The cooperation will support upgrades and new projects at Samsung Electro-Mechanics Sejong and Busan sites, and by 2028 Nvidia is expected to secure a substantial share of the companys package-substrate capacity. Nvidia is also supporting capacity expansion at substrate suppliers including Unimicron Technology Corp. through capacity-reservation agreements.

  Position: AI chip package substrates and AI compute infrastructure supply chain.

  Impact: AI infrastructure capacity expansion is accelerating, significantly expanding the package-substrate supply chain, particularly for ABF substrates. Korean and Taiwan-based substrate suppliers are becoming more deeply integrated into Nvidias supply chain, with 2028 capacity being locked in well in advance. AI compute delivery bottlenecks are extending from chip manufacturing into package substrates.

  Sources: ITHome; Sina Finance
• 2

  MetaX 9M Revenue Expected to Rise 105%–151%; Domestic GPU Maker Turns Profitable for the First Time

  Event: On October 19, MetaX Integrated Circuits (Shanghai) Co., Ltd. issued its earnings forecast for the first nine months of 2026. Revenue is expected at RMB 2.541bn–3.106bn, up 105.60%–151.29% YoY, while net profit attributable to shareholders is expected at RMB 587mn–717mn, turning positive from a year-earlier loss. The simultaneous growth in revenue and profit marks the entry of domestic GPUs into a profitability-at-scale phase as AI compute demand expands.

  Position: Leading domestic general-purpose GPU designer covering AI training and inference workloads.

  Impact: Domestic GPUs are scaling shipments and achieving profitability, moving domestic substitution in AI compute into an earnings-realization phase. Demand for large-model inference and training is driving domestic AI accelerator shipments, while supporting coordinated growth across the domestic GPU ecosystem and related supply chains including advanced packaging and HBM.

  Sources: MyDrivers; MetaX earnings forecast
• 3

  HYGON Q3 Revenue Reaches RMB 5.856bn, Up 45.46% YoY, as Domestic CPU Momentum Continues

  Event: On the evening of October 19, HYGON INFORMATION TECHNOLOGY CO., LTD. (HYGON), a leading domestic CPU supplier, released its Q3 2026 results. Q3 revenue reached RMB 5.856bn, up 45.46% YoY, while net profit attributable to shareholders reached RMB 1.048bn, up 37.90%. Revenue for the first nine months totaled RMB 14.955bn, up 57.58% YoY, while net profit attributable to shareholders exceeded RMB 2.8bn, maintaining strong growth.

  Position: Leading domestic designer of x86-architecture CPUs and DCUs (co-processors).

  Impact: Domestic CPU/DCU penetration continues to rise in servers and AI compute. Chinas IT application innovation (Xinchuang) initiative and AI are jointly driving strong earnings growth, while domestic compute chips are moving from merely usable toward more competitive real-world performance, supporting demand across domestic motherboard, memory and server system supply chains.

  Sources: Securities Times; HYGON Q3 report
• 4

  SK hynix Weighs Nasdaq IPO for Solidigm, Triggering South Korean Dual-Listing Regulatory Controversy

  Event: On October 22, securities-industry reports said SK hynix is evaluating a Nasdaq listing for its U.S. flash-memory subsidiary Solidigm and is considering Morgan Stanley and Goldman Sachs as lead underwriters. Lee Chan-jin, Governor of South Koreas Financial Supervisory Service, unusually commented that the plan "could fall within the scope of dual-listing regulation" and called for the process to be halted. A South Korean lawmaker said the listing day would be "the end of the Korean stock market." Market attention is focused on potential regulatory penalties and their practical enforceability, while verbal intervention by financial regulators could force the company to withdraw the listing plan.

  Position: Enterprise SSD and storage-solutions segment; U.S. subsidiary of SK hynix.

  Impact: If completed, a Solidigm IPO could become a major U.S.-listed semiconductor offering and prompt a revaluation of the enterprise SSD business. The South Korean "dual-listing" controversy highlights tension between cross-border capital-market activity and domestic regulation, while capital restructuring and business carve-outs among global memory leaders remain a key trend to watch.

  Sources: Sina Finance; Toutiao
• 5

  Upstream Materials Breakthrough for Silicon-Based Quantum Chips: Silicon-28 Silane Specialty Gas and Epitaxial Wafers Unveiled for the First Time

  Event: On October 22, the State Key Laboratory of Silicon and Advanced Semiconductor Materials and the Quantum Information Silicon Team at the ZJU-Hangzhou Global Scientific and Technological Innovation Center unveiled two achievements for the first time: silicon-28 silane specialty gas and silicon-28 epitaxial wafers. Silicon-28 isotope purity exceeds 99.99%, filling a critical upstream materials gap for silicon-based quantum chip R&D. The team is led by CAS Academician Yang Deren and Professor Yu Linwei, and the work represents an important milestone in the national-level research project "Preparation and Application of Silicon-28 Isotope Epitaxial Wafers for Quantum Information."

  Position: Core upstream materials for quantum computing; silicon-28 isotope materials form the foundation for silicon-based quantum chips.

  Impact: The domestic breakthrough in silicon-28 isotope materials provides a materials foundation for large-scale fabrication of silicon-based quantum chips based on spin qubits. Semiconductor materials are extending into quantum applications, while domestic supply-chain autonomy in high-end isotope materials is accelerating.

  Sources: Xiaoyu News; ZJU-Hangzhou Global Scientific and Technological Innovation Center

Table 1 · Key Metrics Across the HBM Base-Die Supply Chain (NVHBM Path)

Table 2 · 9M 2026 Earnings Overview of Leading Domestic Compute-Chip Companies

Note: Specific figures are subject to official disclosures by the respective sources. In the event of discrepancies involving qualifications or compliance information, announcements by the competent authorities shall prevail.