Global Semiconductors: HBM, CoWoS, and On-Device AI
HBM supply shortage, advanced packaging competition, on-device AI demand โ the three defining currents in semiconductors right now.
Semiconductors are the most direct beneficiary of the AI era. Three currents are reshaping the industry: HBM memory, CoWoS advanced packaging, and on-device AI. TSMC's foundry monopoly and NVIDIA's GPU monopoly are the central bottlenecks in AI infrastructure supply.
Market Size
Global semiconductor market: $700B+ projected for 2026 (15% YoY growth driven by AI demand)
Key Trends
HBM Supply Shortage
HBM3e demand far exceeds supply. SK Hynix supplies 80%+ to NVIDIA, securing strong bargaining power.
CoWoS Packaging Bottleneck
TSMC's CoWoS packaging capacity shortage is the key bottleneck limiting GPU shipments.
On-Device AI
Growing AI inference on smartphones/PCs stimulating demand for Qualcomm Snapdragon, Apple M-series, Intel Core Ultra.
2nd Foundry Race
Samsung 3GAE and Intel 18A challenging TSMC monopoly, but yield gaps at mass production persist.
Key Players
The semiconductor value chain consists of 7 stages: design โ EDA/IP โ foundry โ equipment โ packaging โ memory โ distribution. In the AI era, the strategic importance of foundry and memory stages is at a peak.
Segments
Logic (CPU/GPU/ASIC)
~45%NVIDIA GPUs are central to the AI era. Hyperscaler proprietary ASICs (Google TPU, Amazon Trainium) growing fast.
Memory (DRAM/HBM/NAND)
~30%HBM is a critical component in AI GPUs. SK Hynix, Samsung, and Micron in a three-way structure.
Analog/Power/Sensors
~25%Growing demand for data center power management semiconductors. AI server power efficiency requirements.
Value Chain
NVIDIA, AMD, Qualcomm, Broadcom
Intensifying AI accelerator design competition
Synopsys, Cadence, Arm
Chip design tools/IP oligopoly
TSMC, Samsung, Intel
TSMC 3nm monopoly is the key bottleneck
ASML, Applied Materials, Lam
EUV equipment export control variable
TSMC CoWoS, ASE, Amkor
CoWoS bottleneck limiting GPU shipments
SKํ์ด๋์ค, Samsung, Micron
HBM supply shortage continues
Geopolitical risk (US-China tensions), power infrastructure, and foundry capacity cycles are the key macro variables for semiconductors.
AI Demand Explosion
GPU/HBM demand structurally exceeding supply on a sustained basis
US-China Semiconductor Conflict
A100/H100 export ban limiting China revenue. Huawei's in-house Kirin chip rising
Foundry Capacity Cycle
TSMC 2nm investment ongoing. CoWoS capacity expansion expected to ease bottleneck in H2 2026
On-Device AI Adoption
Stimulating smartphone replacement cycle โ Qualcomm/Samsung/Apple benefiting
Legacy Semiconductor Inventory Correction
Non-AI semiconductor (MCU, commodity DRAM) demand slowdown widening earnings divergence
The AI semiconductor supercycle is likely to extend through 2027โ2028. However, stock divergence is sharp โ the higher the AI exposure, the greater the benefit; legacy semiconductor exposure increasingly left behind.
Opportunities
- Sustained HBM supply shortage โ prolonged SK Hynix and Micron benefit
- CoWoS capacity expansion beneficiaries (TSMC, Hana Microdisplay) in H2 2026
- On-device AI cycle launch โ Qualcomm, Apple chip segment benefiting
- Custom ASIC market growth โ Broadcom, Marvell benefiting
- Sustained semiconductor equipment demand โ ASML EUV, AMAT, Lam Research
Risks
- NVIDIA valuation at extremes โ earnings miss impact would be severe
- China in-house AI chips (Huawei Ascend) as medium-long term threat
- CoWoS supply expansion easing bottleneck โ near-term premium evaporates
- Legacy semiconductor inventory adjustment (MCU, commodity DRAM) potentially prolonged
Verdict
Semiconductors are the most direct AI era beneficiary. But not all semiconductors ride the same wave. Select stocks based on AI exposure (HBM, CoWoS, GPU design, on-device), and verify AI transition progress before approaching companies with heavy legacy semiconductor exposure.