Paul Krugman argues that surging demand for compute is slowing and may have peaked, suggesting AI's economic impact could be more modest than initially feared. This shift comes as businesses reassess AI investments after companies like OpenAI and Anthropic raised user costs, forcing users to question whether frontier models deliver sufficient value for the price.

Companies abandoned 'tokenmaxxing' strategies when faced with higher user costs, discovering that many AI applications weren't worth the expense. Meanwhile, AI labs struggled to meet user demand, with reports of Anthropic reducing Claude's capabilities during peak periods and limiting access to new models due to computing constraints.

The emerging dynamic mirrors the Industrial Revolution's pattern, which economic historian Robert Allen calls 'Engels' Pause' - where output per worker rose steadily but living standards stagnated for decades. Capital captured outsized gains while workers faced high financing costs that constrained investment in housing and public works.

If AI transforms the economy, the required investment in data centers, power plants, and chip fabs will be unprecedented. However, compute rationing by price should keep workers cost-competitive longer than expected, even as wage pressure builds and inequality grows. The real threat to the AI boom may be social and political backlash rather than economic failure.