AI-driven capital expenditure surprises and mounting anxiety about industry winners and losers have injected renewed volatility into equity markets, Goldman Sachs said, warning that the current uncertainty is unlikely to clear quickly.
The bank pointed to a pronounced bifurcation in recent market performance: memory stocks have rallied while software equities have weakened significantly. Goldman noted that, at the same time, near-term earnings across parts of the sector have remained resilient. Nevertheless, investors are increasingly concentrating on the risk of longer-term disruption as AI adoption spreads.
Goldman expects hyperscaler investment to stay elevated, citing consensus forecasts that put hyperscaler capex at $667 billion in 2026, a 62% increase on the prior year. The firm nonetheless cautioned that the pace of growth will not accelerate indefinitely and that a peak in the growth rate is expected later this year.
Analysts led by Ryan Hammond wrote that "a deceleration in the quarterly growth rate is likely in late 2026," and said this slowdown could expose some AI infrastructure companies. The team highlighted that both revenue growth and valuations in parts of the group appear vulnerable if spending momentum weakens.
Goldman also pointed to the divergence between Nvidia's (NASDAQ:NVDA) share price and its earnings as an example of how hard it can be for companies to sustain strong returns amid investor concerns about "over-earning." The bank argued that as investment shifts from infrastructure to applications, outcomes across equities will become more differentiated.
To identify durable beneficiaries, Goldman said investors will need increasingly granular assessments of competitive positioning, barriers to entry, and pricing power. The note underscored the challenge of using near-term earnings to refute long-term disruption fears, stating: "The concerns about AI disruption will be difficult to disprove in the near-term."
The analysts added that firms facing disruption concerns will probably require either multiple quarters of demonstrated business resilience or materially lower valuations before large-scale investor re-engagement occurs. Goldman emphasized that the recent software sector sell-off was driven principally by long-term uncertainty rather than by a deterioration in near-term earnings.
Looking ahead to 2026, the bank expects more companies to begin quantifying AI-driven productivity gains. However, Goldman cautioned that converting those gains into lasting earnings improvement will depend on whether firms can retain the incremental benefits instead of passing them on through lower prices or higher wages.
Goldman's analysis frames a near-term market environment in which strong headline capex figures coexist with elevated uncertainty about which companies will deliver sustainable value from AI. That combination, the bank suggests, helps explain the recent volatility and the widening performance gap between different subsectors of the technology complex.
Summary
Goldman Sachs warns that AI-related capex surprises and rising disruption concerns are contributing to equity market volatility. While hyperscaler spending is forecast to rise sharply to $667 billion in 2026, a deceleration in growth could pressure parts of the AI infrastructure complex. Near-term earnings appear resilient, but long-term uncertainty is driving differential performance across memory, software, and application-focused stocks.
Key points
- Hyperscaler capex is projected at $667 billion in 2026, up 62% year over year, supporting continued investment in AI infrastructure.
- Market performance has split recently, with memory stocks rallying while software shares have declined, reflecting divergent investor expectations.
- Investors will need detailed analysis of competitive positioning, barriers to entry, and pricing power to identify companies likely to benefit sustainably as AI adoption shifts from infrastructure to applications.
Risks and uncertainties
- Deceleration risk - Goldman warns a slowdown in quarterly capex growth could occur in late 2026, potentially exposing AI infrastructure revenues and valuations.
- Valuation risk - Elevated prices relative to earnings, highlighted by the dislocation between Nvidia's share price and earnings, could hinder sustained returns if growth disappoints.
- Translation risk - Productivity gains from AI may not translate into durable earnings upside if companies pass benefits to customers through lower prices or to workers through higher wages.