NVIDIA's Rubin GPUs Expected to Cut Inference Costs by 90% by Late 2026

haider1 · x · 2026-08-02

The author highlights upcoming trends in AI infrastructure, noting that most of OpenAI's planned data centers are yet to be built. On the hardware front, NVIDIA's Rubin GPUs, expected in Q4 2026, could slash inference costs by around 90%. Furthermore, OpenAI's in-house chips are slated for launch this year and could be 50% cheaper for inference than Vera Rubin.

Original post →

More from Infra

Infra channel →