Fiber optics can carry enormous amounts of data across a network. Yet inside an AI data center switch, information may still travel several inches through copper before it ever reaches the fiber.
That short journey is becoming surprisingly expensive.
The next challenge in optical networking isn’t necessarily making light travel faster. It’s reducing how far electricity must travel before becoming light.
And solving it could change how network switches are built.
The Electrical Bottleneck Before the Fiber
For decades, networking equipment has relied on pluggable optical transceivers.
These small modules sit at the front of a switch. When one fails, a technician can remove and replace it without replacing the entire system.
That modularity helped make high-speed networking practical and serviceable.
But there is a catch.
The switch ASIC typically sits deeper inside the equipment. Its electrical signals must cross circuit-board traces and connectors before reaching the optical modules.
At higher signaling speeds, those electrical paths become harder to manage.
Engineers compensate using advanced SerDes equalization, improved circuit-board materials, and signal-processing circuitry.
The copper connection still works. But maintaining signal quality can require more power and increasingly sophisticated engineering.
This is the hidden cost of keeping optics at the front panel.
Why Moving Optics Closer Changes Everything
Instead of continuously improving that electrical journey, engineers are exploring a different solution.
Shorten it.
Near-Package Optics (NPO) places optical engines close to the switch chip. Co-Packaged Optics (CPO) integrates them within a shared package architecture.
The objective is straightforward: convert electrical signals into light before they travel across longer board-level connections.
Broadcom’s Tomahawk 6 Davisson provides a concrete example.
Announced in 2025, the 102.4-Tbps CPO switch platform uses integrated optical engines. Broadcom reports approximately 70% lower optical-interconnect power consumption compared with traditional pluggable solutions.
NVIDIA is pursuing a similar architectural shift through its Spectrum-X Ethernet Photonics platform.
But moving optics closer creates new challenges.
Front-panel modules are easy to replace. Optical engines integrated near expensive switch silicon require different approaches to packaging, cooling, testing, and maintenance.
And pluggable optics are still advancing. Suppliers including Coherent and Lumentum continue developing 1.6T modules for AI networks.
This isn’t simply a race between old and new transceivers. It’s a decision about where electrical signaling should end and optical transmission should begin.
Pluggable optics made networking modular.
AI is forcing engineers to reconsider the power and complexity required to preserve that modularity.
The next breakthrough may come not from making light faster, but from making electricity travel less.
CONTINUE THE RESEARCH
TECH VISION MAGAZINE · AUGUST 18, 2026
Traditional Optical Architectures: The Dominance and Limits of Pluggable Optics Before the AI Era
Explore pluggable transceiver architectures, SerDes, DSPs, PCB signal loss, optical engines, and the engineering trade-offs driving the transition toward NPO and CPO.