SANTA CLARA — Nvidia has committed at least $6.5 billion to companies developing photonics technology since March, aiming to scale AI infrastructure through optical connectivity. The investments include $2 billion in Lumentum, Coherent, and Marvell, $500 million in Corning, and participation in a $500 million Series E funding round for startup Ayar Labs.
Photonics uses light instead of electricity to transmit data, offering a more energy-efficient alternative for moving information between graphics processing units, memory, networking chips, servers, and data centers. Electrical data transfer consumes more energy, a growing concern as AI deployment expands.
"When you look upstream, you come to the conclusion that we're starting to scale our silicon photonics technology," Nvidia CEO Jensen Huang said at GTC in March. He said the company was beginning to add photonics to its GPU-to-GPU interconnect technology and pointed to its ethernet networking platform used to connect AI factories and GPU clusters. "Which means the amount of silicon photonics technology capacity that we need is substantially higher than the world has today. So we work with the supply chain to make sure we can help them build up the capacity in advance of that," he said.
Nvidia has already incorporated some photonics technology into its networking solutions and announced tools it said would enable AI factories to connect millions of GPUs across sites while reducing energy consumption and operational costs.
"Photonics represents a way for Nvidia to scale their AI infrastructure without the energy costs that staying with electrical and copper will incur," Alvin Nguyen, senior analyst at Forrester, told CNBC. "By investing in photonics companies, Nvidia is making sure that advancements in photonics continue and it will prevent them from hitting a scalability and performance wall that will occur if they remain on electrical and copper," he said.
Still, challenges remain in scaling production. "The technology is sound, production scale is the harder problem," Nick Patience, AI lead at the Futurum Group, told CNBC. "Manufacturing yield on complex co-packaged optical assemblies remains a challenge because precise alignment of optical and silicon components is unforgiving, and when something goes wrong in the packaging process, the assembly typically can't be reworked," he said. "So the transition is underway, but it's still early. I would expect us to see large-scale adoption from 2028 onwards," he added.