Future Trends of 800G and 1.6T Optical Networking: What Network Architects Need to Know

Optical networking is in the middle of its fastest rate transition since 100G went mainstream. 800G is shipping at volume today. 1.6T is entering the lab-to-field pipeline, with first commercial modules expected inside 18 months. For network architects planning data center, DCI, and telecom infrastructure, understanding where each technology fits — and what drives the timeline — is the difference between building a network that scales and one that needs a forklift upgrade in three years.

Where 800G Stands Today

800G optical transceivers are no longer an early-adopter bet. As of mid-2026, multiple tier-1 hyperscalers have standardized on 800G QSFP-DD and 800G OSFP form factors for their newest AI cluster and DCI builds. The ecosystem has matured across three distinct reach categories:

The inflection point for 800G adoption was not technology readiness — it was switch platform maturity. Tomahawk 5 (51.2T) switches with 800G ports entered volume production in 2024, giving network operators a native 800G switching fabric. Before that, 800G optics were being tested in breakout mode (2×400G), which limited the economic case. Today, a full 800G leaf-spine fabric with native 800G switch ASICs is the default architecture for new AI clusters exceeding 10,000 GPUs.

1.6T: What Is Coming and When

1.6T (1.6 Terabit) optical transceivers represent the next major rate jump — doubling 800G. Unlike the 400G→800G transition, which largely reused existing PAM4 technology (just doubling lanes from 4×100G to 8×100G), 1.6T requires new electrical and optical building blocks. The core enablers are:

• 224 Gbps PAM4 electrical lanes. 1.6T modules use 8×200G electrical lanes (OSFP-XD or QSFP-DD800 form factors) with 224G PAM4 SerDes. This is the critical silicon milestone — 112G SerDes enabled 800G; 224G enables 1.6T.

• 200G-per-lane optical channels. For DR8-class 1.6T (8×200G optical), EML and SiPh modulators must support 200 Gbaud PAM4 with sufficient extinction ratio. Early lab results are promising, with multiple vendors demonstrating 200G/lane EML and SiPh prototypes at OFC 2025.

• Coherent 1.6T for DCI. 1.6T ZR/ZR+ coherent modules will use ~140+ Gbaud with probabilistically-shaped 64QAM or higher-order constellations. The DSP complexity roughly doubles versus 800G ZR+ — and power consumption is the primary engineering challenge.

1.6T Timeline Estimates (Industry Consensus, Mid-2026)

Three Technology Shifts Reshaping 800G and 1.6T Optical Networking

1. Silicon Photonics Moving From Niche to Mainstream

At 800G and above, silicon photonics (SiPh) is no longer an alternative to InP and EML — it is becoming the default platform for coherent pluggables and high-volume PAM4 modules. SiPh integrates the modulator, photodetector, and passive waveguides on a single CMOS-compatible die, reducing assembly cost and improving yield at scale. For 1.6T, SiPh is the leading candidate for the optical engine because it can monolithically integrate 8 transmit and 8 receive channels with low per-channel cost. InP EML will retain a role in long-reach and high-output-power applications, but the volume economics increasingly favor SiPh.

2. Linear Drive (LPO) and Co-Packaged Optics (CPO)

As 800G module power consumption approaches 14–18W and 1.6T pushes toward 25–30W, power becomes the binding constraint — not bandwidth. Two approaches are competing to address this:

• Linear-drive pluggable optics (LPO): Remove the DSP from the module and rely on the switch ASIC's SerDes to drive the optical interface directly. Cuts module power by ~30–50% but requires tight co-engineering between switch and optics vendors.

• Co-packaged optics (CPO): Integrate the optical engine directly onto the switch package substrate, eliminating the pluggable interface entirely. Maximum power savings but minimum flexibility — you cannot swap optics without swapping the switch. Currently in trials with hyperscalers; broad adoption unlikely before 2028.

For most network operators, LPO is the pragmatic mid-term path: it retains the operational flexibility of pluggable optics while addressing the power challenge. CPO is a longer-term architectural shift that makes sense only at extreme port densities where pluggable thermal limits are truly exhausted.

3. Coherent Pluggables Extending Reach Without Sacrificing Density

800G ZR+ coherent modules have demonstrated that you can put a full coherent DSP, tunable laser, and optical front-end into a QSFP-DD — collapsing what used to be a 2RU transponder shelf into a switch port. At 1.6T, the same form factor will carry a more complex DSP, and the reach-power trade-off becomes the central design variable. Early 1.6T coherent modules are targeting 120–500 km with EDFA amplification, comparable to 800G ZR+ reach — but achieving this at acceptable power (~25W) requires more advanced DSP node processes (5nm or below) and integrated SiPh front-ends.

Selection Guidance: How to Plan Your 800G and 1.6T Migration

1. If you are building now: Standardize on 800G QSFP-DD or OSFP. Both form factors have clear 1.6T migration paths. Do not lock into a proprietary form factor that lacks an MSA roadmap to 1.6T.

2. If you are planning for 2027: Budget for 1.6T DR8/FR4 modules at roughly 1.5–1.8× the cost of 800G equivalents in year one, trending to ~1.2× by year two. Price parity between 800G and 1.6T per bit is expected around 2028–2029.

3. For DCI links under 120 km: 800G ZR+ is the cost-optimized choice today. 1.6T coherent will arrive in 2027–2028 but at a significant power premium — only justified if your fiber count is genuinely constrained.

4. For AI cluster interconnects: 800G today, 1.6T by 2028. The GPU-to-network bandwidth ratio continues to increase with each GPU generation, and 800G per GPU will be the minimum by the time Rubin and next-gen architectures ship.

5. Power budget first, bandwidth second. A 1.6T module drawing 28W in a 32-port switch is nearly 900W of optics alone — before the ASIC, fans, and PSU losses. Plan your rack power distribution for the 1.6T era now.

Applications Driving the 800G and 1.6T Transition

The demand pull comes from four directions:

• AI/ML training clusters. GPU-to-GPU east-west traffic grows super-linearly with cluster size. 100,000-GPU clusters (operational by 2027) will consume over 100 Tbps of optical interconnect bandwidth.

• Hyperscale data center fabric. Cloud providers are standardizing on 800G spine switches. 1.6T becomes the next logical upgrade when 102.4T switch ASICs ship.

• Metro and regional DCI. 800G ZR+ coherent pluggables are replacing transponder shelves for links under 500 km. 1.6T coherent extends that to higher-capacity links with fewer fiber pairs.

• Telecom transport networks. 400G and 800G coherent wavelengths are the new baseline for OTN/DWDM backbone upgrades, with 1.6T wavelength trials underway at several tier-1 carriers.

APEX GROUP — www.apexallinone.com