A $40 Million Bet That AI Datacenters Can Stop Running Out of Lasers

A tiny compound-semiconductor laser sits inside nearly every fiber-optic link that moves data between chips in an AI datacenter. That laser is almost always made from indium phosphide, and it is running short worldwide (Light Reading). Now the industry is standardizing on an architecture that needs more of them per cluster, intensifying the pinch.

The Open Compute Interconnect (OCI) MSA, a consortium backed by major chipmakers and the largest cloud and AI operators, has converged on wide-and-parallel dense wavelength division multiplexing (DWDM) as the scale-up fabric for AI clusters. Wide-and-parallel DWDM packs many light "colors" onto a single fiber, which means more laser sources, not fewer, per link (Electronics Weekly).

Santa Barbara startup Quintessent is trying to sidestep the bottleneck by building its laser out of gallium arsenide (GaAs) instead. This week the company closed an oversubscribed $40 million Series A led by Cycle Capital, with Goldman Sachs, Ciena, and others joining, and began customer sampling of a single-chip quantum dot DWDM comb laser (Business Wire). The chip generates eight precisely spaced wavelengths from a single device with one bias control, with no banks of individually tuned lasers and no high-power pump lasers. It is heterogeneously integrated on standard silicon photonics for wafer-scale manufacturing (Converge Digest).

Quintessent is the first publicly named startup sampling a non-InP, single-chip DWDM comb laser (Independent.com). Whether the architecture holds at wafer-scale volume, lands real customer commitments beyond sampling, and out-competes the InP-based alternatives the industry is already scaling is the open question the new funding is meant to answer.