According to data from Goldman Sachs analyzed by venture capital firm a16z, large language model prices fell by approximately 95 % between March 2023 and September 2026, dropping from a baseline index of 100 to about 5. A service that cost one dollar at the start now costs roughly five cents. This decline mirrors the drop in personal computer prices over 15 years during the ICT investment wave of the 1980s. The Silicon Data LLM Token Expenditure Index hit a record low of $0.97 per million tokens in August 2026, down 29 % from July and more than 50 % below its peak of $2.05 set in May 2026. Meta, Microsoft, Amazon, and Alphabet face overcapacity risk as massive AI infrastructure investments could see their profitability questioned if demand fails to keep pace with supply.
Source: Read the original article
