A wireless experiment in the middle of a Swiss airfield just told you something about $MRVL's DCI roadmap.

New Nature Comms paper: ETH Zurich + Polariton (the plasmonics startup Marvell acquired in April) ran a 226 GHz sub-THz link over 1,400 m. Net rate 153 Gb/s. The receiver? A plasmonic Mach-Zehnder modulator taking the 226 GHz carrier directly on its electrodes.

No silicon modulator can do this. No TFLN modulator can do this. TFLN's EO response rolls off past ~100 GHz because RF-optical phase matching breaks down. Plasmonic slots don't have that problem: >1 THz measured bandwidth, µm-scale footprint vs cm-scale, fJ/bit vs pJ/bit.

To be precise: this link targets 6G RAN backhaul, not datacenter interconnect. But that's exactly why it matters as evidence. The modulator's THz-class bandwidth just survived a 1.4 km outdoor field trial, not a probe station. That's the same headroom Marvell bought for 3.2T/6.4T coherent ZR/ZR+ and scale-across DCI.

Everyone else scaling coherent baud rates is fighting physics inside silicon and TFLN. Marvell now owns the one modulator platform with bandwidth to spare, integrated with its own DSP and switching silicon.

I keep saying: watch the components, not the announcements. Full breakdown of the plasmonics moat for paid subscribers. $MRVL

$LITE $COHR

Paper: doi.org/10.1038/s41467-…