Why current LLM costs are not sustainable (aditya.patadia.org)

The AI community is facing a significant cost crisis as companies struggle with soaring expenses related to large language models (LLMs). Major firms like Uber have rapidly exhausted their AI budgets, prompting tech giants such as Microsoft and Salesforce to implement spending cuts. High operational costs, exemplified by OpenAI's GPT 5.5 charging $5 per million input tokens and $30 per million output tokens, highlight the unsustainability of current pricing. As model performance improvements plateau and competition increases, it is projected that the costs of these frontier models will eventually drop. Various factors contribute to this potential price reduction, including the emergence of open-weight models that undercut traditional offerings; for instance, GLM-5.2 is significantly cheaper than GPT 5.5 while outperforming it in coding tasks. Additionally, advancements in specialized AI chips could lead to a decrease in inference costs, making operations more efficient. The development of local models also holds the promise of shifting workloads away from costly cloud solutions, empowering users to run applications directly on local hardware. Overall, these dynamics signal a future where competition and technological innovations will drive down costs, ultimately benefiting consumers.