Rubin chips ease AI energy crunch

Nvidia’s Vera Rubin chip claims to deliver the same model work with far fewer chips (1/4 for training) and at far lower inference cost (1/10), promising lower electricity and rack density per unit of AI output. If realized at scale, Rubin could materially reduce the marginal power demand of new data centers and change siting, permitting and grid‑capacity planning. — Lowering per‑workload compute and energy costs shifts the politics of AI (permits, industrial policy, grid planning and climate tradeoffs) by making continued AI expansion more economically and politically defensible.