6 hours ago
20
Nvidia just dropped what might be the most important piece of AI infrastructure that nobody outside of data center engineers will ever see. The Spectrum-6 Ethernet switch, part of the company’s SN6000 series, is now being installed at major AI factories, bringing 102.4 terabits per second of switching capacity on a single chip.
What Spectrum-6 actually does
The Spectrum-6 is the sixth generation of Nvidia’s Spectrum-X Ethernet platform. It supports 200G and 800G ports while integrating co-packaged optics, a technology that embeds optical components directly into the switch package rather than using separate pluggable modules.
Nvidia’s Spectrum-X Ethernet Photonics variants claim five times better power efficiency per port compared with conventional pluggable optics. They also promise five times longer flap-free uptime, meaning fewer of those fun moments where network connections randomly drop and your multi-million-dollar training run crashes.
The switch is designed to manage networks connecting hundreds of thousands of GPUs across both single data centers and distributed environments.
Nvidia also introduced its Spectrum-XGS extension, announced back in August 2025, which enables what the company calls “scale-across connectivity.” This lets distributed data centers, potentially separated by significant distances, function as a single unified giga-scale AI super-factory.
The ecosystem taking shape
The Spectrum-6 was revealed at CES 2026 as part of the broader Rubin platform launch, alongside the BlueField-4 data processing units and ConnectX-9 SuperNICs. Together, these components form the networking stack that Nvidia envisions powering multi-trillion-parameter AI models.
Meta and Oracle deployed earlier Spectrum-X switches in 2025. Nebius has received Spectrum-6 hardware. Cisco is integrating the Spectrum-6 silicon into its own N9100 series switches. Dell and CoreWeave round out the partner roster.
Why this matters for investors
The Spectrum-6’s power efficiency improvements are particularly relevant in a world where AI data centers are approaching gigawatt-scale power consumption. Nvidia explicitly references gigawatt-scale AI super-factories in its positioning. Any technology that delivers meaningful power savings at that scale translates directly into reduced operational costs for hyperscalers.
The partnership with Cisco adds a distribution channel dimension. Cisco’s enterprise sales force is vast, and having Spectrum-6 silicon inside Cisco-branded switches gives Nvidia access to customers who might not buy Nvidia-branded networking gear directly but will purchase it through a trusted vendor relationship.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·