Marvell demos 2nm optical interconnects for 3.2T AI data center links at ECOC 2026
The chipmaker is showcasing 38 technology demonstrations in Málaga, including the industry's first 2nm 400G-per-lane optical PAM4 solution
Marvell Technology just pulled back the curtain on what it calls industry-first 2nm optical interconnect technologies, designed to keep pace with AI data centers that are hungry for bandwidth speeds most of us can barely conceptualize. The showcase is happening at the European Conference on Optical Communication (ECOC) 2026, running September 20 through 24 in Málaga, Spain.
The headline demo: a 2nm 400G-per-lane optical PAM4 solution built to support the jump from 1.6 terabit to 3.2 terabit link speeds.
What Marvell is actually showing off
The company is running 38 technology demonstrations at ECOC, spanning optical PAM4, coherent links, and the silicon that powers them.
Among the standout pieces is an 800G ZR/ZR+ pluggable module. What makes it notable is the integration of MACsec security, powered by Marvell’s Libra DSP. MACsec is an encryption standard for network traffic at the hardware layer, which means data gets protected before it ever leaves the module rather than relying on software further up the stack.
Marvell is also demonstrating a 2nm 1.6T ZR solution alongside coherent-lite technologies.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Perhaps the most ambitious demonstration is a 102.4T co-packaged optics (CPO) platform running 200G-per-lane silicon photonics. Co-packaged optics places optical components directly alongside the switch silicon rather than connecting them through separate pluggable modules. The result is shorter electrical paths, lower power draw, and the ability to push aggregate bandwidth to levels that traditional architectures struggle to match.
Why 2nm matters for optical networking
Every node shrink reduces the energy required per transistor switching event. For optical interconnects, which are deployed by the thousands inside a single data center, even modest per-unit power savings compound into massive reductions at the facility level.
Marvell’s roadmap has been building toward this moment. The company previously delivered 3nm-based 1.6T solutions, and the Electra DSP for 1.6T and Libra DSP for 800G were both announced earlier in 2026. Both products are expected to begin sampling in the second half of this year.