This might be the most important comment from $DELL earnings: “Inference has passed training and is pure demand on our industry.” Dell expects AI inference demand to grow 87X by 2030 to 3,600 quadrillion tokens. (Whaaa???) Training demand? That’s just 5X. And Dell expects enterprise agentic AI to become the single largest workload by 2028. This is why I think people calling for the AI infrastructure buildout to peak are missing the bigger picture. Training was just phase 1. Inference is going to be phase 2. As agents begin performing dozens or hundreds of tasks autonomously, token consumption and compute demand could explode. This is exactly what @daniel_koss has been saying. The AI infrastructure boom isn’t ending when the models are trained…this is when the real consumption starts.