NVIDIA commits $6.5B+ to photonics betting light replaces copper inside AI data centres
Category: AI Infrastructure
What it is
Since March 2026, NVIDIA has invested at least $6.5B into photonics companies:
$2B each into Lumentum and Coherent for silicon photonics and laser components,
and up to $3.2B into Corning for optical fibre factories that could replace the 5,000
copper cables inside rack-scale systems like Vera Rubin. AMD, Alphabet, and
Microsoft have made parallel photonics bets. The thesis: moving data as light
instead of electricity cuts power consumption 5–20× per bit transferred – a critical
unlock for gigawatt-scale AI factories.
Why it Matters for Enterprises
Energy cost and data-transfer bandwidth are the two hidden ceilings on AI scale.
Photonics directly attacks both. Enterprises planning multi-year AI infrastructure
investments should track co-packaged optics timelines - this technology will
determine the cost-per-token floor of the next generation of AI data centres.
