Stanford study: small AI models beat cloud LLMs in 81% of tasks
If this is true, the hyperscalers are toast
A Stanford team compared small language models (SLMs) run on local PCs with cloud-based LLMs across chat and reasoning tasks from 2023 to October 2025. On average, SLMs matched or beat LLMs in 81.2% of cases, with energy and compute costs 50–85% lower. The author argues this could make hyperscaler data-centre investments obsolete.
If this research is true and this trend continues, data centres may be the worst investment in the AI space one can make right now.