In 2024, forecasting AI spend was easy: you just multiplied users by message volume. In 2025, that math didn’t just break—it evaporated. At Palo Alto Networks, we watched our AI spend surge 5.2x year-over-year, and it wasn’t because of a hiring spree. It was “Complexity Variance.”
The shift from simple chatbots to autonomous agents has introduced a new, volatile Bill of Materials. We’re seeing Reasoning Models hit with a 16x unit cost multiplier and Vector Search “Infrastructure Taxes” that dwarf the actual cost of the LLM. In this session, we’re moving past the “Price per Token” myth. We will share the forensic methodology we used to categorize traffic by complexity and show you how to build a 2026 budget that won’t leave you explaining a massive overage to your CFO.