AI inference costs may exceed model training expenses

Enterprises face rising operational costs from AI inference, experimentation, and GPU underutilization. Goldman Sachs estimates global AI infrastructure spending could reach $4 trillion to $8 trillion by 2031. Token pricing varies dramatically across models, making cloud consumption difficult to forecast at scale.