Nvidia pours $6B into photonics startups to slash AI chip energy use
Light-based chips could cut AI energy by 90%—Nvidia bets $6B on it
Nvidia is making a massive bet on photonics, a chip technology that replaces copper electrical connections with light-based data transmission. Over the past three months, the company has invested $2 billion each into three established photonics developers—Lumentum, Coherent, and Marvell—alongside smaller stakes in additional startups, as reported by CNBC. The total investment exceeds $6 billion, signaling that Nvidia sees photonics as critical to overcoming one of AI's most stubborn bottlenecks: the energy and heat generated by moving data across chips.
Current AI chips rely on copper interconnects that consume enormous power as data volumes grow, especially in large-scale training clusters. Photonics promises to reduce energy consumption by up to 90% while enabling faster data transfer rates. For Nvidia, which dominates the AI chip market but faces rising power costs and thermal limits, this technology could allow it to scale future GPU designs without hitting physical constraints. The investments are strategic: Lumentum and Coherent are leaders in optical components, while Marvell brings expertise in data infrastructure. If successful, photonics could redefine how AI hardware is architected, making models like GPT-6 or beyond economically viable to train and deploy.
- Nvidia invested $2B each in Lumentum, Coherent, and Marvell for photonics technology
- Photonics uses light instead of electricity for chip interconnects, cutting energy use by up to 90%
- Copper interconnects are a major bottleneck limiting AI chip scaling; photonics could enable faster, cooler, and more efficient data centers
Why It Matters
Photonics could solve AI's energy crisis, enabling larger models with dramatically lower power costs.