Source: Techmeme
The real signal here isn’t capacity constraints—it’s that AI infrastructure economics have fundamentally inverted: success now creates immediate friction rather than scaling advantage, forcing companies to actively degrade user experience just months after launch, which suggests the current compute-per-inference model is economically unsustainable at mainstream adoption levels and will eventually favor either massive vertical integration (like OpenAI’s Microsoft partnership) or radical efficiency breakthroughs over pure capability races.