NVIDIA Wants Everyone To Rethink AI TCO, and Explains Why “Cost Per Token” Is the Only Metric That Matters

NVIDIA said AI infrastructure TCO should be judged by cost per token, not FLOPS per dollar or raw GPU specs.

The company argues that enterprise buyers should focus on real inference output and how cheaply AI systems generate tokens.

Comparisons between Hopper and Blackwell inference results showed Blackwell delivering much higher throughput and lower token cost.

The framing could shift AI hardware buying decisions toward operating efficiency and real-world inference economics over peak compute specs.