“While the cost of AI inference is being reduced by factors of 10x to 100x, end-user costs are not decreasing because market demand immediately shifts to newer, higher-quality models that consume these efficiency gains.”