Goldman Sachs: AI Capex Boom May Not Have Peaked Yet
Taylor Wilson
Goldman strategist Tony Pasquariello argues that hyperscaler AI capital spending may not yet be at peak growth, with consensus forecasting roughly 80% year-on-year growth through the rest of 2026 — the inflection point may not arrive until Q1 2027, meaning the AI spending tailwind for S&P 500 earnings is far from exhausted.
Has AI capital spending actually peaked?
Goldman strategist Tony Pasquariello states clearly: the current period does not necessarily represent the peak growth rate of the AI capex cycle.
Market consensus projects hyperscaler (Amazon AWS, Microsoft Azure, Google Cloud) capital expenditure growing roughly 80% year-on-year through the remaining quarters of 2026.
This means → even at already elevated growth rates, Goldman sees room for further acceleration or sustained high spending — the top is not yet confirmed.
Why does this number matter for the entire market?
Hyperscalers and AI infrastructure companies are expected to account for roughly two-thirds of S&P 500 earnings growth in 2026.
In plain terms = whether the S&P 500 keeps climbing this year depends, about two-thirds of the way, on these giants' spending plans.
This reflects how concentrated U.S. equity earnings growth has become around the AI supply chain — if capex pivots, the impact extends well beyond tech stocks.
When does the inflection point arrive?
Forecast data points to a clear deceleration starting in Q1 2027.
But Pasquariello emphasizes that the market has repeatedly underestimated the actual scale of hyperscaler investment commitments — the inflection could come later than expected.
This means → the core uncertainty for investors is not "will spending slow?" but "when will it slow, and by how much?"
What does this mean for investors?
Pasquariello frames AI infrastructure spending as the key variable determining the next phase of the AI trade.
Investors must judge whether this boom can continue delivering above-trend earnings growth — if yes, current valuations hold; if not, elevated multiples lose their fundamental anchor.
In plain terms = the question is no longer "is AI useful?" — it is "does the money being poured into infrastructure earn back enough to justify the stock prices?"
Content is for reference only, not financial advice.