Epoch AI reports that the costs associated with AI performance are decreasing more rapidly than those of any previous technology. The "price of thought"—the cost for an AI model to achieve a specific level of performance—drops by about 50 percent every quarter, which represents an annual decrease multiplied by thirteen. This rate of decline surpasses that of historically transformative technologies such as DNA sequencing or lithium batteries. However, this rapid decrease in cost presents significant economic challenges for AI companies, as it makes it harder to retain customers and maintain profitability from large investments. The promise of substantial profits is fading, as seen in the collapsing prices of AI tokens due to increased competition and reduced production costs. The proliferation of freely accessible Chinese AI models and strategic price reductions has created deflationary pressure on the market. While this trend benefits end-users, it casts doubt on the enormous profits that industry players had anticipated. Large investments in AI infrastructure depend on these profit expectations. The LLM Token Expenditure Index, a key indicator from Silicon Data that tracks daily AI token prices, crossed a downward threshold, standing at 97 cents per million tokens on September 1, 2026. On September 28, Epoch AI released a report stating that "the cost of AI is decreasing at an unprecedented rate," outpacing the decline rates of other transformative technologies like DNA sequencing and lithium batteries. To illustrate the speed of this change, Epoch AI compares it to a hypothetical scenario: would you be surprised if the price of a new car dropped from $50,000 to $296 in two years? That is the pace at which AI is becoming cheaper. For example, in January 2025, OpenAI's o3 model achieved a 75 percent success rate on the GPQA Diamond benchmark at an average cost of 30 cents per question. Eighteen months later, the GPT-5.6 Luna model achieved the same score for just 0.04 cents per question, representing a 725-fold drop in the "price of thought" in less than 18 months. This unprecedented deflation creates a major strategic dilemma for leading AI labs like OpenAI and Anthropic, as well as for developers of AI applications. Analysts note that it is increasingly difficult for these companies to ensure customer loyalty or generate strong recurring revenues when prices drop so quickly and cheaper alternatives constantly emerge. Traditional subscription models lose their appeal when users know they can access more powerful and much cheaper services in a short time. For companies investing hundreds of billions of dollars in cutting-edge infrastructure, maintaining sustainable profitability is a complex challenge, as the advantage of a flagship model may be temporary. Despite the rise of generative AI and the launch of ChatGPT nearly four years ago, OpenAI has yet to demonstrate a viable business model.