Nvidia links AI factory value to tokens and watts

Nvidia VP Ian Buck says AI factory economics now center on tokens per watt rather than raw chip performance. Speaking at the Fully Connected event, Buck described tokens as appreciating, revenue-generating assets and noted that Blackwell achieved a 30x improvement in tokens per watt. He added that inference does not replace training, as organizations continually refine deployed models. Nvidia's Groq 3 LPX accelerator pairs with its Vera Rubin platform for latency-sensitive workloads in sectors like fintech, while partners such as CoreWeave help customers balance throughput and token speed.