Chart 01 — The AI Bottleneck Cascade
Every Year, a New Bottleneck. 2026 Belongs to Optics.
How AI infrastructure scarcity has shifted across the value chain · 2024 → 2025 → 2026 → 2027
2024
🖥️
GPU Shortage
NVIDIA H100 / Hopper chips had 6–12 month lead times. Hyperscalers queued billions in orders.
✓ Resolved
2025
🧠
HBM Memory Shortage
High Bandwidth Memory (HBM3/3e) for AI chips fell short. SK Hynix, Micron scrambled to expand.
✓ Easing
2026 ← NOW
💡
Optical Components
800G transceivers 40–60% below demand. 200G EML chips — supply deficit of 25–30%. Lumentum is the only at-volume supplier.
⚡ Active Bottleneck
2027–28
⚡
Power & Cooling
AI factories requiring gigawatts of power. Grid capacity, cooling infrastructure, and energy sourcing emerge as the next frontier.
Emerging
Why This Pattern Matters for Investors
Each bottleneck creates a multi-year growth window for the companies that can relieve it. NVIDIA captured the GPU bottleneck. HBM benefited SK Hynix and Micron. The optical bottleneck of 2026 belongs to Lumentum and Coherent — the only companies with the fabrication capability to ship 200G EML chips at the volume hyperscalers need. McKinsey projects this shortfall persists through 2027 at minimum.
CURIOUS INVESTING INSIGHTS