Since March 2026, Nvidia has committed at least $6.5 billion to companies developing photonics technology. This isn’t a small side bet — it’s a strategic wager on how AI data centers will work in the future.
Why photonics?
The problem is straightforward: data needs to move between GPUs, servers, and data centers. Today, that happens over copper cables using electrical signals. At the data volumes that AI training and inference generate, copper hits physical limits — heat, energy consumption, speed.
Photonics replaces copper with light. Fiber-optic connections can move more data, faster, and more efficiently. For AI data centers that need to network millions of GPUs, that’s a decisive advantage.
Where the money is going
Nvidia made three major investments — roughly $2 billion each:
- Marvell Technology: Strategic partnership for silicon photonics
- Lumentum: One of the leading manufacturers of photonic components
- Coherent: Developer of photonic technologies for data centers
Jensen Huang put it plainly at GTC in March: the world doesn’t have nearly enough silicon photonics capacity for what’s coming.
What this means for AI
When Nvidia — the company that essentially built the GPU infrastructure of the AI era single-handedly — puts $6.5 billion into a technology, that’s a signal. Photonics isn’t becoming relevant someday. It’s being built now.
If you’re wondering where the next bottleneck in AI development sits: it’s not the models, not the data, not even the chips. It’s the question of how to move data between chips fast enough. And Nvidia is betting that light is the answer.
Sources: