The current gold rush into artificial intelligence may eventually hit a wall, according to Jan Hatzius, the chief economist at Goldman Sachs. Speaking at the firm’s Communacopia and Tech conference, Hatzius cautioned that while the surge in spending has been historic, nothing moves upward on a chart indefinitely. He warned that investors who assume this boom will persist forever could face significant challenges when the tide inevitably turns.
Hatzius explained that most technological revolutions follow a predictable cycle consisting of an intense build-out phase followed by an exploitation phase. In the first stage, companies spend aggressively to establish infrastructure and develop capabilities. Once those tools are in place, however, the focus shifts toward using the technology efficiently, which naturally leads to a decline in overall investment volume. Even under his own optimistic baseline assumption—that AI is productive and will drive long-term economic growth—a slowdown remains inevitable.
There is also a darker possibility that looms over the sector. Hatzius noted that it is entirely possible many of these massive investments will prove unproductive, creating a downside scenario that analysts simply cannot ignore. This skepticism comes despite staggering industry forecasts, such as recent figures from PwC suggesting global AI infrastructure spending could reach 31.6 trillion dollars by 2050. These projections envision data center capital expenditures ballooning from around 800 billion dollars annually in 2026 to nearly two trillion by mid-century.
Currently, tech giants like Microsoft, Google, and Meta are pouring billions into chips and hardware to fuel their ambitions. Industry leaders argue that because internet-connected equipment requires frequent upgrades every few years, the demand for capital will only accelerate. However, as Hatzius suggests, the transition from building a foundation to extracting actual value often creates volatility for those who mistook a temporary spike for a permanent trend.