完全な半導体サプライチェーンマップ 2026

The semiconductor supply chain is the most complex and strategically important industrial network in the world. In 2026, it spans from a single company in the Netherlands that makes lithography machines to trillion-dollar cloud providers consuming millions of chips per quarter.

The Equipment Layer

At the very foundation sits ASML, the sole manufacturer of extreme ultraviolet (EUV) lithography systems. Each machine costs over $350 million and requires 100,000 components from 5,000 suppliers. Without ASML, no advanced chip can be made. Applied Materials and Lam Research provide complementary deposition and etch tools.

The Foundry Bottleneck

TSMC controls roughly 60% of global foundry revenue and over 90% of advanced node production (5nm and below). This single-point dependency means that any disruption in Taiwan would cascade through every sector of the global economy. Samsung Foundry and Intel Foundry Services are attempting to compete but remain years behind.

The Design Layer

NVIDIA dominates AI accelerator design with over 80% market share in data center GPUs. AMD is the primary challenger with MI300X. Broadcom has carved out a lucrative niche designing custom AI chips for Google (TPU) and Meta (MTIA). Marvell and Marvell compete in custom silicon and optical interconnects.

Memory: The Hidden Bottleneck

High Bandwidth Memory (HBM) has become as important as the GPU itself. SK Hynix leads with roughly 50% market share in HBM3E, followed by Samsung and Micron. Every H100 GPU requires 80GB of HBM, and the Blackwell B200 requires 192GB - creating severe supply constraints.

The Demand Side

Microsoft, Google, Amazon, and Meta are spending over $200 billion combined on AI infrastructure in 2026. This capex flows through the entire chain: from GPUs to memory to foundries to equipment makers. Understanding this flow is the key to anticipating which stocks move when earnings surprise.

Investment Implications

The most actionable insight from mapping this chain is identifying which companies have the highest "cascade sensitivity" - that is, which stocks move most reliably when a leader reports earnings. Our data shows that memory stocks (SK Hynix, Micron) react within 24 hours of NVIDIA earnings, while equipment names (ASML, AMAT) take 2-5 days to fully price in the implications.