May 13, 2026

1.6T Optical Modules: The AI Interconnect Bottleneck Breaker (Q3 2026 Deployment Guide)

Q3 2026 production. Lead times now 32 wks.

  • 1.6T deployments production: Q3 2026
  • Module shipments: 80K Q1 2026 to 1.2M Q4 2026 (15x)
  • CPO port cost: $2,800-$3,400 vs pluggable $4,200
  • SiPh 800G-DR4: $480/port May 2026 (down from $720 Q3 2025)
  • EML 200G/lane yield: 38% Q2 2025 to 70% Q1 2026
  • GB200 NVL72 aggregate bandwidth: ~130 Tbps per rack
  • Broadcom Tomahawk 6: 102.4T, sampling Q4 2026
  • 1.6T-DR8 ASP drop: $6,100 to $4,200 in 14 months
  • OSFP-XD power: 25W pluggable vs 15W CPO per 1.6T port
  • Marvell DSP share at 800G/1.6T: ~58%
  • EML laser capacity Q3 2026: ~18M units vs ~24M demand
  • 1.6T-DR8 lead time May 2026: 26-32 weeks
  • CPO share of 1.6T+ ports end-2027: 12-18% (Dell'Oro)
  • Photonic AI processor 2026 volume: <5,000 units combined
  • LPO power savings: 50% vs DSP-based pluggable

Answer First

Hyperscale cloud providers — led by Microsoft, Meta, Google, and Amazon — are pushing 1.6T optical module deployments into production in Q3 2026, roughly 18 months earlier than the original 2027-2028 industry roadmap. The trigger: AI training clusters built around NVIDIA GB200 NVL72 and AMD MI355X racks now require 1.6 Tbps per port to keep GPUs from idling on east-west traffic.

Pricing Curve (May 2026 actuals)

| Module | Q3 2025 ASP | Q1 2026 ASP | May 2026 ASP | |---|---|---|---| | 400G-DR4 | $720 | $580 | $510 | | 800G-DR4 (EML) | $1,450 | $1,180 | $980 | | 800G-DR4 (SiPh) | $720 | $560 | $480 | | 1.6T-DR8 | — | $6,100 | $4,200 | | 1.6T CPO port | — | $3,400 | $2,800 |

Bottom Line

1.6T is a Q3 2026 production story driven by GPU rack bandwidth, enabled by EML yield breakthroughs, and supply-constrained by laser fabs and DSP availability.

Subscribe to aiinframap Weekly

One email Friday morning. Engineers only.