A report from Deutsche Bank, published on October 14, 2026, indicates that the cost of achieving a specific level of AI performance has decreased by nearly 47% every quarter since 2023. This trend results in a 13-fold annual reduction in costs.
The bank compared AI cost deflation to other technologies. AI costs are falling four times faster than DNA sequencing, six times faster than general compute, and 18 times faster than lithium-ion battery costs.
Data from Epoch AI shows a 725-fold cost reduction for matching scores on the GPQA Diamond benchmark, comparing early 2025 OpenAI models to subsequent versions.
Despite falling unit costs, capital expenditure by US and Chinese hyperscalers on AI infrastructure is projected to reach $1 trillion in 2026. Experts estimate that AGI systems may require 1,000 times more compute than currently available.
The report identifies verification and energy as significant constraints. Reliability and verification expenses are rising, while power limitations could hinder productivity growth before AI reaches superintelligence.
