AI capex is outrunning the revenue it's meant to buy

Every earnings call this quarter had the same shape: capex up again, AI revenue still "early." The gap between what hyperscalers spend on compute and what customers actually pay for inference keeps widening. I'm not a bear — I use these tools daily — but eventually the depreciation math has to meet the P&L. What I want to know is what people are seeing on the ground: are enterprise budgets actually moving from pilots into production, or is this the year of the quietly renewed pilot? Real numbers would be more useful than vib