Inference prices are falling faster than budgets are growing

Our team's per-token inference bill dropped roughly 70% over the last year, yet total AI spend went up. We just keep routing more work to models because it got cheap enough to justify. Anyone else seeing this? It feels like the opposite of normal procurement: falling unit prices are expanding our budget rather than shrinking it. Curious how others are forecasting this for 2025 planning, and whether finance teams are catching on to the fact that cheap tokens are not the same as a cheap AI program.