The surging costs associated with AI computing, particularly for "tokens" (the basic unit of AI measurement), are prompting major companies to re-evaluate their AI strategies. Some enterprises have exhausted their annual AI budgets within months or seen their spending double or triple. This has led to efforts to ration AI use, direct employees toward less expensive internal tools, and improve skills to optimize returns. For instance, Uber reportedly blew through its annual budget for "agentic" or autonomous AI use by March, and Google processes over 3.2 quadrillion tokens monthly, a seven-fold increase year-over-year. Microsoft has limited employee access to certain advanced AI programs, and Salesforce has implemented systems to track token use against business outcomes. Critics point to this cost-conscious shift as a potential slowdown for AI growth and a concern for companies like Anthropic and OpenAI, which are nearing public listings. However, some investors and executives believe sales and usage growth continue to outpace forecasts.
Companies are also facing a dilemma with the soaring costs of advanced AI models. For example, only 18% of spending on advanced AI coding tools translates to shipped products, according to EntelligenceAI, highlighting a need for improved model efficiency. Executives at companies like Meta are emphasizing that AI tool usage must contribute to productivity, not just be used for the sake of it. Instances of employees using powerful, premium-tier models for simple queries further underscore the inefficient use of resources.
This cost pressure is leading some businesses to explore more affordable options. While major AI developers like Anthropic, OpenAI, and Google offer cheaper versions of their flagship models, many companies are also considering lower-cost AI models, including those developed in China. Chinese open-source AI models, such as Kimi K3 by Moonshot AI and DeepSeek’s latest V4, are gaining global attention for being more affordable and are now seen as serious competitors. Kimi K3, for example, is touted as the world’s largest open-source AI model with 2.8 trillion parameters, and its recent public rollout experienced such high demand that new subscriptions were temporarily halted due to overwhelmed capacity.
The rise of affordable Chinese AI models is putting pressure on American technology titans, with concerns that these models could undercut the pricing power and demand for U.S. AI companies, despite U.S.-led restrictions on China's access to advanced chip technology. The surge in demand for Kimi K3 highlighted the challenges Chinese AI models face in scaling their infrastructure to meet a growing user base, both domestically and internationally. Moonshot AI acknowledged the unexpected popularity of Kimi K3, stating that demand pushed against their capacity limits and they are moving to add more capacity.
This landscape underscores a significant shift towards a mixed-model approach, where companies combine different AI models, including more affordable, open-source, or Chinese-developed options, to manage costs and optimize their AI strategies. This evolving environment is set to redefine the economics and influence of key players within the booming AI industry.
