1.6T Optics: The AI Bottleneck Breaker

Hyperscalers just pulled the 1.6T optics timeline forward by 18 months. Microsoft, Meta, Google, and Amazon are deploying 1.6T-DR8 in production starting Q3 2026.

The trigger

A single NVIDIA GB200 NVL72 rack pushes ~130 Tbps east-west. At 800G that needs 162 ports per rack. At 1.6T it's 81. Half the radix pressure, ~35% less power per bit.

The enabler

EML laser yield at 200G/lane PAM4: 38% Q2 2025 → 70% Q1 2026. That dropped 1.6T-DR8 ASP from $6,100 to $4,200 in 14 months.

Three architectures

**Pluggable 1.6T-DR8** — $4,200/port, 25W, 26-32 wk lead time. **CPO** — $2,800-$3,400/port, 15W, no field replace. **SiPh 800G-DR4** — $480/port, 22% undercut vs EML 800G.

Reverse-hype

CPO will only hit 12-18% of 1.6T+ ports by end-2027 (Dell'Oro). Pluggables dominate through 2029. Watch LPO: 50% less power, 30% less cost.