NVIDIA Debuts Silicon Photonics Switches for AI Data Centers

NVIDIA has unveiled a pair of co-packaged silicon photonics networking switches that it says will allow AI facilities to connect millions of GPUs across sites while drastically reducing energy consumption and operational costs. Company CEO Jensen Huang introduced the switches at Intel's GTC 2025 event in Silicon Valley.

“AI factories are a new class of data centers with extreme scale, and networking infrastructure must be reinvented to keep pace,” Huang said. “By integrating silicon photonics directly into switches, NVIDIA is shattering the old limitations of hyperscale and enterprise networks and opening the gate to million-GPU AI factories.”

The photonic/electronic fusion technology features in two distinct product lines. Both offerings, one Ethernet and one InfiniBand, increase data transfer speed to 1.6 Tb/s per port, enabling the operation of millions of GPUs. The products additionally use 4x fewer lasers to deliver 3.5x more power efficiency, 63x greater signal integrity, 10x better network resiliency at scale, and 1.3x faster deployment compared to traditional methods, NVIDIA said.

NVIDIA has released two networking switches based on silicon photonics technologies. According to the company, the technologies will enable AI companies to connect millions of GPUs. Courtesy of NVIDIA.

NVIDIA Spectrum-X Photonics Ethernet switches include multiple configurations, including 128 ports of 800 Gbps or 512 ports of 200 Gbps, delivering 100 Tbps total bandwidth as well as 512 ports of 800 Gbps or 2048 ports of 200 Gbps, for a total throughput of 400 Tbps. Spectrum-X Photonics Ethernet switches are expected in 2026.

NVIDIA Quantum-X Photonics InfiniBand switches provide 144 ports of 800 Gbps and use a liquid-cooled design to efficiently cool the onboard silicon photonics. NVIDIA Quantum-X Photonics switches offer 2× faster speeds and 5× higher scalability for AI compute fabrics compared with the previous generation. The product is planned for release later this year.

NVIDIA's silicon photonics ecosystem includes TSMC, Browave, Corning Incorporated, Fabrinet, Foxconn, SENKO, SPIL, Sumitomo Electric Industries, and TFC Communication. Additionally, the switches integrate technologies from laser developers Coherent and Lumentum.

“Innovations that drastically lower networking power consumption will expand the market for photonic components, supporting larger AI installations and accelerating the transition from copper to photonic interconnects,” said Wupen Yuen, Lumentum president for cloud and networking technology and product.

Lumentum said that the new laser technology complements its existing electro-absorption modulated lasers (EMLs), which power many of today’s 400Gb/s, 800Gb/s, and upcoming 1.6Tb/s optical transceivers in cloud and AI data centers. "Photonics switch technology will revolutionize data centers and advance a new wave of AI factories," said Gilad Shainer, senior vice president of networking at NVIDIA. "NVIDIA’s collaboration with innovators such as Lumentum will enable the next-generation of million scale AI."