Why does it matter?
A large annual token bill would turn inference efficiency into a central business cost.
Direct quote
More significantly, these agents will execute over 217 billion actions per day and consume 3.7 TeraTokens/Calls (3,700,000,000,000) daily to support this still rapidly expanding inferencing load. The token delivery cost worldwide for supporting all these agent actions will surpass $68 billion annually, but the cost to complete an ever more complex and sophisticated individual action will be 87% lower.
IDC