THE PRICE OF AI IS QUIETLY COLLAPSING AND NOBODY CAN FIND THE FLOOR The price of running artificial intelligence is falling fast and has not found a bottom. The Silicon Data LLM Token Expenditure Index, which tracks what companies actually pay per unit of AI output, dropped nearly 20 percent from its May high and has not recovered. Bloomberg flagged the decline last week, and the timing is uncomfortable for an industry trying to justify the largest infrastructure investment cycle in technology history. The longer-term picture is worse. Over the past 18 months, the cost to run a frontier AI model has fallen somewhere between 80 and 150 times depending on the workload and provider. LLM API prices dropped roughly 80 percent between early 2025 and early 2026 alone. Budget models now start at $0.10 per million input tokens. Even frontier reasoning models have seen prices compress under competitive pressure from every direction. The forces driving this are structural. Better model architectures require less compute for the same output. Open-weight models that companies run without paying per-token fees put a ceiling on what closed providers can charge. More GPU supply comes online every quarter. For AI startups on per-query revenue, margin is being crushed. For cloud providers with hundreds of billions committed to GPU infrastructure, the economics depend on price stabilizing at a level that has not arrived. For investors who funded AI companies at peak valuations, a 20 percent decline in the index they assumed would keep rising is not a comfortable number. Keywords: LLM token price decline, AI pricing collapse, Bloomberg AI index, AI market valuation 2026