Investors are underestimating the AI boom, Goldman Sachs says

Goldman Sachs warns the AI spending surge may be larger than investors expect. The bank projects hyperscaler capital expenditure could reach about $1.1 trillion in 2027 versus roughly $920 billion reflected in current Wall Street estimates, and in a bullish scenario spending could top $1.4 trillion.

Goldman bases its view on accelerating demand for AI computing: it forecasts token consumption will rise 24-fold through 2030, driven largely by enterprise agents. More tokens mean more compute, which pushes demand for data centres, chips, networking equipment and power infrastructure, and higher input costs lift the nominal capex needed to support token growth.

Cloud providers have provided a supporting signal, with Google Cloud and AWS reporting a combined backlog of $832 billion in the first quarter, up from $358 billion six months earlier. Still, the firm flags key risks. Several companies have noted rising token expenses for AI tools, leaving open whether productivity gains will outweigh operating costs.

goldman sachs, ai spending, hyperscalers, capex, token consumption, enterprise agents, data centres, cloud backlog, google cloud, aws