Photonics Infrastructure Attracts Major AI Investments

DollarCollapse reports that photonics, the use of light to carry data inside and between chips, is receiving significant strategic capital as AI data centers scale. Per DollarCollapse, NVIDIA invested $4 billion in March 2026, allocating $2 billion each to Lumentum and Coherent, with the deals including multi-billion-dollar purchase commitments and funding for new U.S.-based fabrication facilities. DollarCollapse also reports Marvell acquired Celestial AI for $3.25 billion, and Nokia acquired Infiner for $2.3 billion.

DollarCollapse outlines the basic photonics building blocks that matter for data-center designs:

• •A laser that converts electrical signals to light
• •A waveguide that routes light on-chip or between components
• •A modulator that encodes data onto light pulses
• •A detector that converts light back to electrical form

These components enable higher bandwidth and lower thermal load than copper at large scale, per the reporting.

Editorial analysis: Companies building large-scale AI clusters face growing bandwidth and thermal limits with copper interconnects, creating demand for optical interconnects and integrated photonics. Observers of comparable infrastructure transitions note that suppliers of specialized components and foundry capacity often capture outsized returns relative to systems integrators.

For practitioners and infrastructure teams, track three indicators: announced purchase commitments and fab buildouts (which indicate capacity timelines), consolidation or M&A among photonics component makers, and emerging standards for on-chip optical interfaces. These signals will influence procurement lead times, integration testing requirements, and cross-vendor interoperability work.

Editorial analysis: As training scales to models that now exceed 100 trillion parameters, optical interconnects and laser/modulator supply chains become material constraints for cluster throughput and power efficiency. That shifts integration risk from GPU procurement alone toward end-to-end optical networking and packaging engineering.