The price of AI is crashing faster than the rate of Moore's Law, report suggests
The price of AI "intelligence," has fallen dramatically in recent years, faster than any other transformational technology in history according to some estimates. This raises serious questions about the need to retain access to frontier capabilities, when they show up elsewhere so swiftly.
A new report by Epoch AI reveals that the cost of artificial intelligence (AI) has plummeted by thousands of times in recent years, faster than any other transformative technology in the past century. The report indicates that AI costs are now about 50% cheaper every quarter, and 13 times cheaper annually. This rate of decline is faster than lithium batteries, DNA sequencing, and even Moore's Law, which has been a constant driving force behind compute performance and efficiency for the past 60 years.
AI costs fell by hundreds of thousands of times within just five years, while compute costs dropped by hundreds of billions of times over 60 years. AI's cost reduction is now accelerating four times faster than DNA sequencing and 18 times faster than lithium batteries. This rapid cost decrease poses a challenge for AI service providers like OpenAI and Anthropic, who are racing for more advanced AI capabilities.
The challenge arises because it becomes increasingly difficult to justify staying with a current provider when a more capable and cheaper alternative emerges in just a few weeks or months. The report's findings are based on models achieving an 81.25% or better score on the Graduate-Level Google-Proof Q&A (GPQA) Diamond benchmark, measuring AI capabilities across various knowledge domains.
However, the rate of cost reduction has been declining, with the initial rate of 66% per quarter dropping to 32% per quarter two years later. This decline could indicate a slowdown in AI progress or increased difficulty in reducing costs, especially considering the rising prices of energy and compute hardware.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
