NVIDIA announced that its Spectrum-X6 networking switches, developed at its Israeli R&D center, will be deployed by top AI infra operators to deliver 102.4 terabits per second of bandwidth, enabling AI factories to operate as end-to-end systems with improved utilization and reduced bottlenecks. The Vera Rubin platform, which combines Vera CPUs and Rubin GPUs with advanced liquid cooling, claims up to a 10x reduction in inference costs versus Blackwell and can cut required GPUs by up to fourfold, with production ramping in 2026 and initial shipments already to hyperscalers like AWS, Google Cloud, and Microsoft. Nvidia also outlined its Vera CPU based on the Olympus microarchitecture, designed for agentic AI workloads with fast per-core performance and low memory latency, using a monolithic die and a high-speed interconnect. Early real-world benchmarks and field tests highlighted dramatic gains in tokens-per-watt and agentic latency, reinforcing Nvidia’s ecosystem-driven approach across hardware, interconnects, and software. Production and partner momentum, including ties to major cloud providers and AI firms, underscore Nvidia’s push to expand beyond GPUs into a broader data-center stack for next-generation AI workloads.