Bank of America's Raedler flags $1 trillion AI capex and a 50% token price drop

Bank of America's note frames the current AI buildout as a capital investment cycle with precedents. Sebastian Raedler examined whether AI infrastructure spending could follow the pattern of earlier cycles, in which expanding capacity eventually placed pressure on profitability. The bank points to rising investment expectations, increased competition and declining prices for AI services as the shared features.

A key contrast in the note is the cloud computing market of the 2010s, where barriers to entry helped limit competition and supported industry profitability. In AI, competition has increased at the model level, including from Chinese developers offering open-weight models (models whose underlying weights are openly released). That difference matters because earlier cycles were often characterised by initially elevated profit margins, which encouraged additional investment and eventually contributed to excess capacity and reduced pricing power.

The bank also weighed whether lower AI service prices could generate enough additional demand to offset the decline in revenue per unit of usage. Whether that demand response is strong enough may determine how the cycle's financial returns play out, a question the note raises rather than resolves.