Nvidia’s Israeli-Developed Spectrum-6 Switch Rolls Out in Major AI Data Centers

Nvidia announced that its advanced Spectrum-6 communication switch, developed at its Israeli R&D center, will be deployed in AI data centers of leading technology companies including Microsoft, SpaceX AI, Tesla, CoreWeave, and Nebius. The switch, part of Nvidia’s Vera Rubin platform, delivers data transfer speeds of 102.4 terabits per second, doubling the speed of the previous generation. Nvidia stated that Spectrum-6 provides the bandwidth, scalability, and intelligence required to operate AI data centers as unified computing systems, enabling cloud providers to offer larger, high-performance computing resources that accelerate AI model training, deployment, and operation.

Training and running large AI models depend on synchronized work among thousands of processors exchanging data continuously. Traditional internet switches are designed for traffic between users, servers, and storage, not for the complex synchronous communication needed between processors in AI data centers. Nvidia’s Spectrum series, developed in Israel, bridges this gap by enabling fast data flow to every processor, allowing more AI processors to function as a single unit and improving computing efficiency for complex tasks. This results in faster processing times and greater large-scale efficiency.

CoreWeave’s network product manager Min Yon highlighted that Spectrum-6 integration supports their infrastructure’s bandwidth, resilience, and efficiency, essential for demanding AI workloads. Nebius Vice President Orly Rossman emphasized that at massive scale, performance depends on coordination among AI chips to prevent bottlenecks, a challenge Spectrum-6 addresses. Nebius adopted the switch early to maintain speed and robustness as customer workloads grow.