NADELLA WARNS AGAINST TOKEN-MAXING, ADMITS HE DOES IT

Nadella warned that deploying the most powerful AI models indiscriminately wastes resources. The productivity gains from using frontier models must justify their token costs, he argued, suggesting companies should match computational power to task complexity. Despite this stance, Nadella confessed to being a "token-maxer" himself. "It is addictive," he said, highlighting the temptation even senior tech leaders face when deploying advanced AI systems. The comment underscores a growing tension in AI adoption: while frontier models like GPT-4 deliver impressive results, their high computational costs make them economically inefficient for simple tasks. Smaller, specialized models often provide better value for everyday applications. Nadella's candid admission suggests awareness of the challenge extends to boardrooms, even as the AI industry continues building increasingly capable—and expensive—models.