Technology Insights · Industry Trends · Product Knowledge · Application Notes · News & Updates
Data center optics have doubled in speed roughly every three years for two decades. But each generation brought a different engineering breakthrough — and a different migration challenge. Understanding what changed at each step helps network architects plan future upgrades without repeating past mistakes. Here is the evolution from 10G to 800G, what each transition cost, and what it teaches about the move to 1.6T.
The Speed Generations at a Glance
| Generation | Year | Modulation | Lanes | Breakthrough | Migration Pain |
|---|---|---|---|---|---|
| 10G | ~2010 | NRZ | 1 × 10G | SFP+ form factor | — |
| 25G | ~2015 | NRZ | 1 × 25G | SFP28 (same SFP+ cage) | Server NIC upgrade |
| 100G | ~2017 | NRZ (4×25G) | 4 lanes | QSFP28, parallel optics | MPO-12 connector adoption |
| 400G | ~2020 | PAM4 (8×50G) | 8 lanes | PAM4 modulation, QSFP-DD | PAM4 signal integrity, MPO-16 |
| 800G | ~2024 | PAM4 (8×100G) | 8 lanes | 112G SerDes, silicon photonics | Power density, fiber count |
| 1.6T | ~2027 | PAM4 (8×200G) | 8 lanes | 224G SerDes, 3nm DSP | Switch power, CPO transition |
The Two Discontinuities That Changed Everything
PAM4 at 400G was the hardest transition. Moving from NRZ (two signal levels) to PAM4 (four levels) cut the signal-to-noise margin in half while doubling the bit rate per lane. Every imperfection in the fiber plant — dirty connectors, tight bends, aging splices — went from manageable to catastrophic. This was the generation where structured cabling stopped being optional and became non-negotiable.
Silicon photonics at 800G was the second discontinuity. Integrating modulators, receivers, and MUX functions on a CMOS silicon die collapsed the cost of optical engines from hundreds of discrete components to a single chip. It is the same wafer-scale economics that drove CPU costs down — now applied to optics. This is why 800G per-gigabit cost is lower than 400G was at launch.
What Each Migration Teaches About 800G→1.6T
Three patterns repeat across every speed transition:
- The form factor stays constant for two generations. SFP+ handled 10G and 25G. QSFP28 handled 40G and 100G. QSFP-DD handles 400G and 800G — and will handle 1.6T before CPO replaces it.
- The cable plant outlasts the optics by 3–5×. Fiber installed for 10G in 2010 is still carrying 800G in 2026. Every dollar spent on quality cabling in the initial build pays back for 15+ years.
- Power density compounds. 10G drew ~1 W per port. 800G draws ~16 W. 1.6T will draw ~25 W. The switch power supply and rack cooling you install today must accommodate the next generation.
- Planning rule: When deploying at generation N, ensure the cable plant, switch power supplies, and rack cooling can support generation N+1. The optics will be swapped in 3–4 years. Everything else should last 10+ years. The margin you build into infrastructure today is the cheapest capacity upgrade you will ever buy.
APEX Group supplies optical transceivers from 10G SFP+ to 800G QSFP-DD ZR+, with the same procurement relationship spanning every speed generation — a single partner from legacy 10G links to next-generation 1.6T coherent.
APEX GROUP — www.apexallinone.com


