Tower Semiconductor (TSEM)

Merchant silicon-photonics foundry (Tower) — the fab behind AI optical engines

AI exposure 37% — derived from reported segment revenue, not asserted.

Where it sits

Silicon Layer → Foundry Services · figures rebased to Q2 2026, reported 2026-08-04

Rule 1 — placed by what it makes. Tower is a foundry: it manufactures customers' designs on differentiated specialty processes rather than designing chips of its own, and every platform it runs is foundry work. The optical, mobile and industrial markets it serves are value-chain edges, not classification tags — the same treatment TSMC receives, whose HPC platform is 66% of revenue and still buckets as Foundry Services.

Reported segments

Tower Semiconductor ($460M Q2 2026, +24% YoY, record; Q3 guided $520M, +31%) — Israeli specialty analog foundry, and the largest merchant silicon-photonics foundry. Tower reports a SINGLE operating segment (semiconductor foundry services); the platform mix below is management commentary from the Q2 2026 call, not segment disclosure. By platform: RF Infrastructure 49% (+140% YoY) + Power Management 14% + RF Mobile 12% (RF SOI -14% YoY, deliberately shrinking) + Sensor & Display 12%, with ~13% unallocated. SILICON PHOTONICS sits inside RF Infrastructure at a >$680M annualised run-rate (+270%…

How the 37% is built

Investment read

Moat. A production process lead in silicon photonics that customers have paid to reserve: $290M of prepayments for capacity and $1.3B of 2027 revenue already contracted, behind a production-ready 400Gbps/lane silicon modulator and more than five million coherent photonic ICs shipped to Marvell. Qualification cycles run 12-18 months like the rest of the optical chain. But TSMC and GlobalFoundries offer silicon photonics directly, and indium phosphide is a competing architecture entirely.

Bottleneck / pricing power. Allocation power without much price power. Capacity is spoken for — customers prepaid $290M to reserve it — yet gross margin is only 30%, a company record but low for a foundry whose output is scarce. Specialty foundry pricing is competed wafer by wafer; the 2028 model targets 45%, and hitting it is what would turn scarcity into pricing.

Role in the AI stack. The merchant specialty foundry behind AI optical engines.

Connected companies

Others in Foundry Services

All 174 companies · What moved this quarter · How the map is built