1. Semiconductor CAPEX: Global Headlines (Sep 1 – Oct 1)
Over the past month, major foundries, OSATs, and material suppliers have accelerated back-end CAPEX expansion, driven by HBM4 transitions, 2nm price competition, and data center buildouts.
- TSMC ramped up CoWoS & SoIC capacity through Q3, signaling long-term structural investment in advanced packaging.
- SK hynix officially began HBM4 preparation, with a focus on yield and capacity expansion.
- Samsung Foundry resumed equipment orders for its Taylor fab in Texas and offered a $20K/wafer price signal for 2nm nodes — interpreted as both a competitive move and a strategic bait for new customers.
- Amkor finalized plans for a large advanced packaging and test campus in Arizona, tightening US-based supply chains.
- Samsung Electro-Mechanics (SEMCO) and LG Innotek both revealed roadmaps for large FC-BGA substrates and glass substrates targeting server/AI demand by 2027–28.
- Hanmi Semiconductor unveiled 2.5D big-die TC/FC bonders at SEMICON Taiwan, optimized for HBM and GPU module bonding.
- Major US hyperscalers pushed data center CAPEX to record levels, accelerating immersion cooling and optical I/O adoption.
👉 Further Reading: Semiconductor Industry Deep Dive｜The 2nm, EUV & Advanced Packaging Inflection
2. CAPEX Shift: Front-End vs. Back-End
For decades, semiconductor CAPEX has been dominated by front-end investments—EUV lithography, GAA transistors, and transistor scaling. But that dominance is shifting.
- Advanced packaging—CoWoS, SoIC, Foveros—has become a second axis of performance scaling alongside lithography.
- TSMC plans to push CoWoS monthly capacity to 75,000 wafers by 2025, a scale intended to support post-HBM4 and Blackwell-generation devices.
- Equipment bottlenecks in hybrid bonding, FC/TC bonding, metrology, and large FC-BGA/glass substrates are shaping investment flows.
- Meanwhile, High-NA EUV adoption is cautious at TSMC (cost-efficiency focus) and aggressive at Intel (14A). This divergence is fragmenting front-end CAPEX distribution across customers and regions.
Interpretation: Back-end CAPEX is no longer a supporting act — it’s becoming the main stage for performance and yield gains.
3. Global Leaders: Equipment, Packaging, Substrates
Lithography & Front-End
- ASML: High-NA EUV tools cost $350–400M each. Adoption is gradual; TSMC focuses on cost optimization while Intel pushes hard.
- Applied Materials: Acquired 9% stake in BESI (hybrid bonding), bridging front-end and packaging.
- ASM International: Lowered H2’25 guidance; ALD demand remains structural but node timing is volatile.
Packaging & OSAT
- TSMC acts as a de facto advanced packaging house, running CoWoS & SoIC at full throttle.
- ASE expects advanced packaging/test revenue to reach $1.6B in 2025 (+160%).
- Amkor’s Arizona facility will be a key hub for US-based advanced packaging.
- JCET is scaling fan-out and CoWoS-like capacity within China.
Substrates & Materials
- Demand for large AI server FC-BGA substrates is surging. Ibiden and Unimicron are expanding; SEMCO and LG Innotek are entering with server-grade and glass roadmaps.
- Ajinomoto (ABF) is planning a 50% expansion by 2030, including US localization.
4. Korea’s Value Chain Position: Equipment · Substrates · OSAT
Equipment
- Hanmi Semiconductor: Leading in big-die TC/FC bonding for HBM and GPU modules. Precision alignment is key for thermal performance.
- Etching, deposition, measurement, and laser equipment makers (e.g., Wonik IPS, Jusung, EO Technics, Park Systems) are benefiting from hybrid bonding and RDL demand.
Substrates
- Samsung Electro-Mechanics (SEMCO) is Korea’s only volume producer of server-grade FC-BGA substrates.
- LG Innotek is preparing MLC and glass substrate mass production by 2027–28.
- Simmtech remains strong in memory module substrates, tightly correlated with AI memory cycles.
OSAT / Testing
- Hana Micron is expanding packaging capacity in Vietnam to diversify from China and align with US supply chains.
- Amkor Korea is expected to play a supporting role in Arizona with domestic material & equipment partners.
🇰🇷 Korea’s edge lies in its memory + equipment + substrate synergy, with location advantages in the US–Vietnam supply chain triangle.
5. Customer Roadmaps: N2, Pricing & Yield Dynamics
⚠️ This section is based on public disclosures and industry consensus — not confirmed roadmaps.
- Apple is likely to be an early N2 adopter for power efficiency, leveraging chiplets for flexibility.
- MediaTek continues to sample 2nm aggressively for premium mobile/edge AI.
- NVIDIA focuses on packaging + HBM4, using dual foundries to optimize lead time and cost.
Samsung’s $20K/wafer 2nm pricing is a customer acquisition tool; TSMC is taking a multi-patterning route without High-NA to control cost.
6. Power & Cooling Infrastructure: A Hidden CAPEX Frontier
- Immersion and liquid cooling are scaling fast in US and EU data centers. Air cooling is no longer sufficient for high-density AI workloads.
- Microfluidics inside silicon can reduce hotspot temperatures by 65%—a game changer for 3D stacks.
- Optical I/O is emerging to lower power and increase bandwidth inside advanced packages.
These infrastructure shifts are tightly coupled with back-end investments: substrates, bonding, and power delivery are now part of the same CAPEX story.
7. Investment Outlook (2026–2028)
1️⃣ Back-End CAPEX Becomes Structural
CoWoS-L, SoIC, hybrid bonding, and large/glass substrates are transitioning from niche to core CAPEX categories.
2️⃣ HBM4 Transition Wave
TSV/interposer specs are rising sharply. Bonding accuracy, substrate flatness, and cooling are key enablers.
3️⃣ Power & Cooling Investment Joins the Party
Immersion cooling, microfluidics, and optical I/O will become CAPEX companions to packaging.
4️⃣ Korea’s Leverage Points
Korean companies can anchor CAPEX flows through memory, substrates, and equipment — positioning themselves as key nodes in the global supply chain.
8. Kori’s Take 📝
The shift from front-end to back-end CAPEX is a migration of bottlenecks.
Performance scaling is no longer just about transistor density; it’s about who can stitch the pieces together.
TSMC and US hyperscalers are moving first, but Korea has a unique chance to position itself as a critical node through memory, substrates, and advanced packaging equipment.
2026–2028 will be decisive: technology, geography, and CAPEX timing will converge.
International Monetary Fund | IMF
9. Q&A
Q1. Why is back-end CAPEX growing so rapidly?
Because front-end scaling alone no longer drives performance. Chiplets, CoWoS, and 3D stacking are now critical, pushing massive investments into packaging, substrates, and bonding equipment.
Q2. Where is Korea’s competitive edge?
In memory (HBM), substrates (FC-BGA, glass), and bonding equipment. Korea also benefits from US–Vietnam supply chain positioning.
Q3. What should investors watch in the next 3 years?
HBM4 transition, hybrid bonding adoption, cooling infrastructure CAPEX, and which companies can solve the bottlenecks.
10. 🇯🇵 日本語SEOまとめ（要約）
2025年以降、**半導体CAPEX（設備投資）**は前工程（リソグラフィ・EUV）から後工程（パッケージング・基板・OSAT）へと構造的にシフトしています。
特にHBM4・CoWoS・SoIC・2nm・グラス基板といったキーワードは、今後3年間の投資サイクルの中核になると見られています。
韓国企業は、HBMメモリ・FC-BGA/グラス基板・ボンディング装置という強みを活かし、米国・ベトナムとのサプライチェーン再編で重要な役割を果たす可能性が高いです。
👉 検索用キーワード:
半導体CAPEX, 後工程投資, HBM4, CoWoS, SoIC, 2nm, グラス基板, 韓国企業, 投資戦略, パッケージング
#SemiconductorCAPEX #AdvancedPackaging #HBM4 #CoWoS #2nm #GlassSubstrate #KoreaSemiconductor #InvestmentStrategy
