The Semiconductor Supercycle: Why MU, AVGO and QCOM Are Structural Longs

Semiconductor cycles used to be inventory cycles: over-order, glut, crash, repeat. The 2026 cycle is different in kind — demand is programmatic (multi-year AI capex plans), supply is consolidated (three memory makers, one leading-edge foundry), and product is co-designed with customers years ahead. That combination converts what used to be commodity cyclicality into contracted, visible earnings.

Three structural longs

• Micron (MU) — the HBM oligopoly member trading at the biggest discount to its new margin structure; the market still prices a cycle that no longer exists.
• Broadcom (AVGO) — custom XPU programs for at least five hyperscalers plus the Ethernet networking franchise; backlog visibility measured in years.
• Qualcomm (QCOM) — the edge-AI sleeper: on-device inference is becoming a handset, auto and PC attach story, and the market still values it like a maturing modem vendor.

What breaks the thesis

Watch Intel's foundry progress (a credible second leading-edge source would deflate pricing power across the chain), China export-control escalation, and any hyperscaler capex deceleration. None are visible in the data today — but supercycle theses die on the day everyone stops questioning them, so we keep the checklist explicit.