Joachim Klement predicts small on-device models will undercut cloud AI spending

Investment strategist Joachim Klement argues that small language models running on local PCs and phones will replace large cloud models, leaving hyperscalers' data-centre spending unprofitable.

He says the AI bubble will burst in 2027 or 2028 and that the S&P 500 could fall from about 7,500 to about 5,000 by end-2027. He cites a May Stanford study comparing small models such as Qwen 3 and Gemma 3 with cloud models including ChatGPT 5 and Claude Sonnet 4.5. The report gives Klement's forecast and his reading of that study; it does not report results from the study itself beyond the models compared.