The S&P 500 Is Over-Earning. The Next Leg Depends on What Replaces AI Capex

Goldman Sachs sees earnings slowing rather than breaking as the extraordinary lift from AI spending, semiconductor margins and private investment gains begins to fade. The handoff now moves from capex toward productivity, with 8,700 still built on earnings growth rather than a richer market multiple.

• S&P 500 earnings are undeniably running hot, but Goldman sees deceleration rather than an earnings collapse, forecasting EPS of $415 in 2027 and $460 in 2028.

• AI capex has supplied an extraordinary earnings tailwind, but rising depreciation means that contribution falls from 11 percentage points in 2026 to a marginal drag by 2028.

• Semiconductor margins are the sharper earnings fault line, and a return toward historical gross margins could remove roughly 10% from S&P 500 earnings.

• Goldman still sees the S&P 500 reaching 8,700 over 12 months, but the engine is earnings rather than multiple expansion, making the quality of the AI productivity handoff increasingly important.

There is a difference between saying the S&P 500 is over-earning and saying the S&P 500 is sitting inside an earnings bubble, and that distinction is the centre of Goldman Sachs strategist Ben Snider and his team’s latest US equity work. Earnings have unquestionably gone vertical by historical standards. S&P 500 EPS rose 51% year on year in the second quarter and 26% over the latest four quarters, a pace exceeded over the past three decades only during the post-recession rebounds of 2010 and 2021. Earnings are running far above their long-run trend and well ahead of what the historical relationship with US economic growth would normally suggest.

But the important word here is overearnings, not bubble.

Goldman’s base case is that earnings growth slows rather than falls off a cliff. The bank forecasts S&P 500 EPS rising 11% to $415 in 2027 and another 11% to $460 in 2028. That view rests on solid economic growth, some easing of current energy pressure, and eventually a transition from the earnings benefits of building AI infrastructure to the productivity benefits of actually using it.

For traders, that changes the question. The market is no longer simply debating whether AI is real. It is debating whether the extraordinary profits generated while everyone builds the runway can survive once the concrete trucks begin leaving the airport.