Amid the flood of news from Nvidia’s annual GTC event, one item stood out. Nvidia introduced new silicon photonics network switches that integrate network optics into the switch using a technique called co-packaged optics (CPO), replacing traditional external pluggable transceivers. While Nvidia alluded to its new switches providing a cost savings, the primary benefit is to reduce power consumption with an improvement in network resiliency.
Pluggable transceivers have been the de facto standard way of connecting switches with optical cables for as long as there has been optical connections. The transceiver modulates light waves that transmit data, making it much more efficient than copper. The integration of this capability removes connection points, improving performance as well as reducing power.
To put this into perspective: A single Nvidia photonics switch replaces 72 traditional transceivers and reduces the number of lasers by 432 per switch. In large-scale AI data centers with thousands of switches, this equates to hundreds of thousands or even millions of transceivers. The use of CPO translates to a 3.5x improvement in power efficiency, 10x greater network resiliency, and 1.3x faster deployment time, according to Nvidia. Given my days as a network engineer having had to connect optics and replace them when they fail, these numbers seem reasonable, although the 1.3x faster deployment time seems low given transceivers are somewhat delicate, and it can be time consuming to deploy these at scale.
Silicon photonics enables Nvidia to build a network for the era of reasoning, providing the capability to support more users, to increase the number of tokens across the data center, and to bring those data centers to market quickly. While most of the focus of AI infrastructure has been on servers and GPUs, the network has become an essential part of the computing systems. AI has driven data centers to evolve into what Nvidia refers to as “AI factories,” which require massive-scale connectivity across thousands of GPUs, making high-speed optical networking more important than ever.
Nvidia’s photonics family of switches include the Spectrum-X and Quantum-X series. The former is built for AI data centers that use Ethernet, while providing significantly more bandwidth than traditional Ethernet setups. Its configurations include 128 ports of 800 Gbps or 512 ports of 200 Gbps, delivering a total bandwidth of 100 Tbps. For larger-scale deployments, Spectrum-X can expand to 512 ports of 800 Gbps or 2,048 ports of 200 Gbps, reaching a whopping 400 Tbps. The Ethernet switches are coming to market in 2026.
Quantum-X is designed for InfiniBand, a high-speed networking technology used for AI model training and inference. It has 144 ports of 800 Gbps and uses 200 Gbps SerDes to boost performance. Quantum-X features liquid cooling, which makes it more energy efficient and keeps it stable under heavy workloads. It’s twice as fast and five times more scalable than previous InfiniBand switches, making it ideal for AI tasks that require communication between GPUs. Quantum-X switches will be available later this year.
Nvidia typically uses partnerships where appropriate, and the new switch design was done in collaboration with multiple vendors across different aspects, including creating the lasers, packaging, and other elements as part of the silicon photonics. Hundreds of patents were also included. Nvidia will licensing the innovations created to its partners and customers with the goal of scaling this model.
Nvidia’s partner ecosystem includes TSMC, which provides advanced chip fabrication and 3D chip stacking to integrate silicon photonics into Nvidia’s hardware. Coherent, Eoptolink, Fabrinet, and Innolight are involved in the development, manufacturing, and supply of the transceivers. Additional partners include Browave, Coherent, Corning Incorporated, Fabrinet, Foxconn, Lumentum, SENKO, SPIL, Sumitomo Electric Industries, and TFC Communication.
AI has transformed the way data centers are being designed. During his keynote at GTC, CEO Jensen Huang talked about the data center being the “new unit of compute,” which refers to the entire data center having to act like one massive server. That has driven compute to be primarily CPU based to being GPU centric. Now the network needs to evolve to ensure data is being fed to the GPUs at a speed they can process the data. The new co-packaged switches remove external parts, which have historically added a small amount of overhead to networking. Pre-AI this was negligible, but with AI, any slowness in the network leads to dollars being wasted.