Goldman Sachs warns the market underestimates AI investment, forecasting 2027 data center capital expenditure at $110 billion, potentially $140 billion, versus Wall Street's $92 billion estimate. Driven by enterprise AI agents, global AI token consumption may rise 24-fold by 2030, fueling demand for infrastructure like chips and power. However, risks include resource shortages, uncertain efficiency gains, and overvalued stocks detached from fundamentals.