The industry has been treating inference as a single market — a commodity measured in cost per token. Jensen just showed it’s five distinct markets, each with different hardware requirements, different economics, and different willingness to pay. When I described it to Stuart as being like T-Mobile’s cell phone tiers — free old iPhone versus premium iPhone 16 Pro with the high data package — he immediately built on it: “Every token is not the same. How do you segment your data center workloads to make sure you’re giving users what they need to deliver maximum value?”