Banks are moving to curb surging token costs as they expand the use of artificial intelligence, after AI pricing models shifted from subscriptions to token-based billing. Executives at several banks have publicly flagged the growing burden. Banks are reassessing which models to apply to which tasks, opting for older or smaller models where possible. They are also considering open-source options, reusing stored answers and expanding in-house computing infrastructure. Some tasks may remain cheaper to handle with people.

China\'s AI companies are rapidly expanding global market share with low-cost open-source models that users can download and modify, as the United States regulates access to high-performance models from OpenAI and Anthropic. Some Chinese systems are assessed as comparable to Anthropic\'s top model in certain cybersecurity scenarios. The U.S. is easing some export bans but maintains a national export-control stance. Companies globally are rolling out new AI chips, tools and systems, while demand grows for cost-efficient AI use.