Nvidia's latest Vera Rubin chip reduces inference token costs by ten times relative to Blackwell, which is particularly relevant in an environment where compute constraints are limiting growth, and demand for chips that optimize compute usage will be strong.