noticias de la compañía sobre Why does NVIDIA need Rubin?

NVIDIA needs Rubin for four big, business-critical reasons—agentic AI scale, system-level performance, cost leadership, and ecosystem lock-in—as AI shifts from single-GPU tasks to massive, always-on “AI factories."

Modern AI (long-context reasoning, multi-step agents, generative video/code) requires 10–100* more tokens and compute than earlier chat models.

• Rubin is purpose-built for million-token contexts and agentic workflowsNVIDIA Corporation.
• It delivers 50 petaFLOPS of NVFP4 inference—far beyond Blackwell’s limitsNVIDIA.
• Without Rubin, NVIDIA could not serve the exponential growth in inference and training demandNVIDIA Corporation.

AI performance is no longer about one fast GPU—it depends on data movement, communication, and full-system efficiencyNVIDIA 开发者.

• Extreme co-design: Six chips (Rubin GPU, Vera CPU, NVLink 6, BlueField-4 DPU, ConnectX-9, Spectrum-6) work as one rack-scale supercomputerNVIDIA Corporation.
• NVLink 6 + HBM4: Eliminates inter-GPU and memory bottlenecks; 1.7 PB/s memory bandwidth per rackNVIDIA Corporation.
• 100% warm-water cooling: Cuts data center energy use and costs dramatically.
• Result: 10* higher efficiency vs. Blackwell; predictable, stable throughput for 24/7 AI factoriesNVIDIA.

Customers now prioritize cost per token over raw FLOPSNVIDIA.

• Rubin cuts inference token cost by ~10* while boosting throughput.
• More tokens per watt: 10* better energy efficiency; a 1GW data center on Rubin can generate **$150B/year** in revenue (vs. $30B for Blackwell).
• This lets NVIDIA keep leadership in price/performance and block competitors.

Hyperscalers (Microsoft, Google, AWS) and AI leaders (OpenAI, Anthropic) are standardizing on Rubin for next-gen supercomputersNVIDIA Newsroom.

• Full-stack platform: From chips to software (CUDA-X, DSX), Rubin creates high switching costsNVIDIA.
• Rack-scale turnkey systems (NVL72/NVL144) simplify deployment for “AI factories"NVIDIA Corporation.
• Confidential computing: 3rd-gen security across full racks—critical for enterprise and government customersNVIDIA.

Rubin is not just a new GPU—it’s NVIDIA’s AI factory operating system for the agentic AI era. Without it, NVIDIA would lose its lead in performance, efficiency, cost, and ecosystem control as AI scales to industrial levels.