Today in AI: Nvidia Vera Rubin vs. AMD Helios — Inference Cost War and Grid Capacity Crisis — July 23, 2026
July 23, 2026 — curated links and takeaways. 1. NVIDIA Vera Rubin Platform Cuts Inference Cost Per Token 10x, Ramps Full Production Nvidia's Vera Rubin NVL72 rack-scale system, now in full production across Taiwan server makers, delivers 10x lower inference cost per token and 10x higher throughput per megawatt versus Blackwell, with Google Cloud A5X … Read more