Companies Shift to Small AI Models Amid Soaring Token Costs

Technology hyperscalers and financial institutions are restructuring their operational and funding strategies to manage the escalating costs of artificial intelligence. To support an estimated $725 billion in capital expenditures for chips and data centers this year, Amazon.com and Alphabet Inc. are executing record-breaking bond sales in non-U.S. currencies, including euros, yen, and sterling.

Simultaneously, enterprises are experiencing sticker shock as AI providers like OpenAI and Microsoft transition from flat subscriptions to usage-based token pricing. This shift has led some companies, such as Uber, to exhaust annual budgets within months. In response, firms are increasingly utilizing routing tools to assign tasks to cheaper open-source alternatives, including Chinese models like DeepSeek, while reserving premium models for complex workloads.

Financial institutions are taking more aggressive steps to decouple from external vendors. PNC Financial Services Group is investing in internal GPU compute and deploying smaller, more efficient language models to prevent token costs from erasing productivity gains. Meanwhile, OpenAI is reportedly considering price cuts to remain competitive against Anthropic as demand for affordable AI options grows. Gartner predicts that AI coding costs could exceed average developer salaries by 2028, further driving the trend toward smaller, cheaper models.