Computing architecture redefined: Nvidia Vera Rubin - SiliconANGLE

The rise of agentic AI is compelling key technology providers to reevaluate computing architectures capable of supporting rapidly expanding autonomous systems. NVIDIA and CoreWeave's unveiling of the ... Read original →

The Vera Rubin NVL72 isn’t just about scaling GPUs—it’s a structural response to agentic AI’s demand for persistent, real-time inference. By integrating 72 GPUs with BlueField-4 DPUs and 260 TB/s NVLink 6, NVIDIA and CoreWeave are forcing co-evolution across cooling (liquid now essential), networking (InfiniBand/RoCE dominance), and semiconductor processes (3nm EUV for power efficiency). Geopolitically, tightening U.S. export controls and reliance on Taiwan, China for advanced packaging heighten supply chain fragility, pushing firms to diversify assembly/test capacity. Competitively, AMD may accelerate MI400 integration with Infinity Fabric, while Intel’s Gaudi 4 lacks the full-stack cohesion of NVIDIA’s Spectrum-X + Quantum-X800 stack. Within 18 months, AI infrastructure will pivot from training-centric to inference-as-a-service—where operational cost per trillion-parameter model defines market leadership.