The data center industry stands at a critical inflection point in optical interconnect technology. With 800G Ethernet emerging as the dominant speed grade for next-generation switch platforms and AI training fabrics, network architects face a fundamental decision: which modulation technology best serves their specific reach requirements, power budgets, and cost targets? The answer is no longer straightforward, as the traditional boundary between Pulse Amplitude Modulation 4-level (PAM4) intensity-modulation/direct-detection (IM/DD) and coherent detection continues to blur at the 800G generation.

Historically, PAM4 and coherent optics occupied distinctly separate application domains. PAM4-based transceivers dominated intra-data-center connectivity at distances up to 10 km, offering simplicity, low power consumption, and aggressive cost scaling. Coherent optics, with their sophisticated digital signal processing (DSP) and modulation of amplitude, phase, and polarization, owned the long-haul and metro transport segments where fiber impairment compensation and spectral efficiency were non-negotiable. However, the development of 400ZR pluggable coherent optics starting in 2020 fundamentally changed this landscape by bringing coherent technology into compact QSFP-DD and OSFP form factors suitable for data center routers and switches. As of 2025, with the OIF 800ZR Implementation Agreement released in October 2024 and multiple vendors shipping 800G coherent pluggables, the overlap between PAM4 and coherent reach envelopes has grown significantly.

This article provides a comprehensive, research-grade comparison of 800G PAM4 and 800G coherent technologies across all relevant data center application regimes. It examines the physics and engineering trade-offs behind each approach, analyzes real-world deployment considerations including power consumption, cost structures, fiber utilization, and scalability, and provides decision frameworks for architects planning next-generation data center networks. The analysis draws on the latest industry standards (IEEE 802.3ck/cu/df, OIF 800ZR, Open ROADM MSA v8.0), market data, and deployment experience from hyperscale and enterprise environments.

2.1 The Journey to 800G: Standards and Industry Milestones

Understanding the current 800G landscape requires tracing the evolution of data center optical interconnects over the past decade. The transition from 100G to 400G, and now to 800G, was driven by successive doublings in switch ASIC SerDes lane rates, which in turn enabled higher aggregate port speeds without proportionally increasing lane counts or module complexity.

At 100G, the industry primarily relied on four-lane solutions using 25G NRZ (Non-Return-to-Zero) signaling per lane. The 100G-SR4, DR4, FR4, and LR4 standards represented a mature ecosystem with well-understood cost and power profiles. The transition to 400G introduced PAM4 modulation as a necessity: rather than scaling to 16 lanes of 25G NRZ, the IEEE 802.3bs standard defined 400G-DR4 and 400G-FR4 using four lanes of 100G PAM4 (53 Gbaud x 2 bits/symbol). This was the first mass-market adoption of PAM4 in data center optics, and it proved that multi-level signaling could work reliably at scale inside the data center.

Simultaneously, the 400G era saw the birth of pluggable coherent for data center interconnect (DCI). The OIF 400ZR Implementation Agreement, finalized in March 2020, defined a single-wavelength 400G coherent interface using DP-16QAM at approximately 60 Gbaud, targeting 80-120 km amplified DWDM links in QSFP-DD and OSFP form factors. This was transformative: for the first time, coherent optics could be plugged directly into routers and switches without requiring dedicated transport equipment. The Open ZR+ Multi-Source Agreement (MSA) extended this concept, enabling flexible rate operation (100G/200G/300G/400G) with adaptive modulation for reaches up to 1000+ km over ROADM-equipped line systems.

The 800G generation builds on both traditions. IEEE 802.3ck (ratified in 2022) defined 800G Ethernet (800GBASE-R) using eight lanes of 100G PAM4 at 106.25 Gbps per lane (53 Gbaud PAM4). The OIF released the 800ZR IA in October 2024, specifying single-wavelength 800G coherent transmission using DP-16QAM at approximately 118 Gbaud for 80-120 km DCI. The Open ROADM MSA v8.0, released in April 2025, further defined 800G ZR+ with interoperable probabilistic constellation shaping (PCS) for extended reach applications. As of 2025, multiple vendors have announced or begun shipping 800G modules in both PAM4 and coherent variants, setting the stage for a complex technology competition across overlapping reach domains.

2.2 Market Context: AI-Driven Demand and 800G Adoption

The explosive growth of artificial intelligence workloads has fundamentally accelerated the timeline for 800G adoption. AI/ML training clusters require massive east-west bandwidth between GPU servers, and the spine-leaf fabric architectures connecting these clusters are rapidly transitioning from 400G to 800G port speeds. According to Cignal AI, the datacom optical component market grew by over 60% in 2025 to exceed $16 billion in revenue, driven primarily by continued growth in 400G and 800G module shipments. Innolight, Coherent Corp., and Eoptolink have emerged as the largest suppliers of datacom modules, while Coherent Corp., Broadcom, and Lumentum serve as key sources of critical optical components.

The 800G transition is not merely an incremental upgrade. AI training workloads generate traffic patterns fundamentally different from traditional cloud computing. GPU-to-GPU traffic inside a pod or across racks can easily reach hundreds of gigabits per second per node. NVIDIA's DGX H100 server, for instance, features 8 H100 GPUs, each requiring two 200G optical connections, totaling 16 transceivers per server and demanding at least four 800G top-of-rack (ToR) switch ports. Scaling to larger training clusters with thousands of GPUs translates directly into thousands of 800G optical modules per deployment.

Sanjay Yadav

Optical Communications & Network Automation Expert | Author of 3 Books for Optical Engineers | Founder, MapYourTech

Optical networking engineer with nearly two decades of experience across DWDM, OTN, coherent optics, submarine systems, and cloud infrastructure. Founder of MapYourTech. Read full bio →

Follow on LinkedIn

Related Articles on MapYourTech

Coherent Optics · Jul 10, 2026

The Submarine Cable Stack: Open Cables, Generalized OSNR, SDM and Power-Constrained Transmission

Coherent Optics · Jul 10, 2026

Coherent DSP Architecture: From ADC Samples to Client Bits

Automation · Jul 10, 2026

Building an Optical NOC Dashboard with OpenConfig Telemetry

Coherent Optics · Jul 10, 2026

Spectral Efficiency in Practice: The Four Levers