The velocity of S&P 500 earnings growth over the past year has unsettled even experienced market observers. But a closer reading of Goldman Sachs Research's Ben Snider and team’s September 17, 2026 report, "Over-earning, but not an earnings bubble", argues that alarm and complacency are both misplaced reactions. The picture is more specific and more instructive than either camp allows.
The Diagnosis: Earnings Are Elevated, Not Illusory
S&P 500 EPS grew 51% year-over-year in Q2 2026 and 26% over the past four quarters. That pace has only been exceeded during the post-recession rebounds of 2010 and 2021. Snider et al. are direct: earnings have "recently outpaced both their historical trend and their historical relationship with US GDP growth." The forward 12-month P/E sits at 19x, matching its 10-year average. That reading looks benign until one notes that the cyclically adjusted P/E has climbed to one of its highest levels on record, surpassing 2021 and falling just short of the peak of the Dot-Com Bubble.
The distinction matters. A market trading at average multiples on inflated earnings is not the same as a market trading at average multiples on durable earnings. The team at Goldman Sachs isolates three specific factors driving the inflation: AI capital expenditure flows, semiconductor margin expansion, and unrealized equity investment gains. Each is real, each is temporary, and each will fade.
The AI Capex Tailwind: Nearly Half of This Year's Growth
The AI investment boom is the dominant force in recent earnings momentum. The mega-cap hyperscalers are on track to spend $800 billion on capex in 2026, a 94% increase over 2025. That spending is flowing through to the earnings of the entire AI infrastructure complex, including semiconductors, tech hardware, industrials, and utilities. According to the report, this complex is "collectively accounting for roughly half of consensus S&P 500 earnings growth this year." AI capex has added an 11-percentage-point tailwind to S&P 500 EPS growth in 2026.
The tailwind, however, is transitional by design. As hyperscaler capex growth decelerates and depreciation expenses on existing AI assets mount, that 11 pp contribution is expected to turn into a 1 pp drag by 2028. The team notes that "if capex were fully expensed, consensus 2027 S&P 500 EPS would be 22% lower than under current accounting rules," a gap that represents one of the widest in recent decades. In an upside scenario where hyperscaler capex reaches $1.5 trillion in 2027, the contribution to EPS growth rises to 13 pp. In a downside scenario at $1 trillion, it falls to 1 pp. The range is wide. The direction of travel, on the baseline, is clearly lower.
Semiconductors and "Other Income": Two More Temporary Boosts
Semiconductor gross margins sit at their highest level in decades, driven by strong demand and constrained supply. Memory firms within the S&P 500 currently carry gross margins near 80%, more than double historical averages. Goldman's analysts estimate that about one-quarter of S&P 500 semiconductor earnings growth in 2026 stems from margin expansion alone. Supply is expected to remain tight through 2027, but the rate of incremental expansion is forecast to slow meaningfully. The report quantifies the sensitivity plainly: each percentage-point change in semiconductor gross margins shifts S&P 500 EPS growth by approximately 1 pp.
Then there is "other income." Mega-cap tech companies booked over $150 billion in unrealized gains on private equity stakes in Q2 2026, lifting S&P 500 earnings by 12%. More is expected in H2 2026, but these accounting contributions "should diminish in 2027." Their complete removal would create an 8-percentage-point drag on 2027 EPS growth relative to 2026. Excluding this income from 2026 entirely, the 2027 EPS forecast would instead reflect growth of 18%.
The Baseline Forecast: Deceleration, Not Collapse
Goldman Sachs forecasts S&P 500 EPS of $415 in 2027, up 11%, and $460 in 2028, also up 11%. These are driven by solid GDP growth, modestly declining energy prices, and a fading but still present boost from AI productivity. Snider et al. embed a 1 pp boost from AI adoption in 2027 and a 2 pp boost in 2028. They are candid that "the impact of AI adoption on corporate earnings still appears narrow," even as enterprise adoption accelerates. The 12-month S&P 500 price target is 8,700, implying a 14% return. The expected driver is earnings growth, not multiple expansion.
Five Key Takeaways for Advisors and Investors
- The "earnings bubble" framing is the wrong lens. Earnings are elevated above trend and above what GDP growth would ordinarily support, but Goldman Sachs forecasts deceleration, not collapse. The base case is 11% EPS growth in both 2027 and 2028.
- AI capex is real but reflexive. Nearly half of 2026 S&P 500 earnings growth traces back to the AI infrastructure spending cycle. As that capex growth slows and depreciation builds, the tailwind reverses. Exposure to AI infrastructure stocks should be assessed with that trajectory in mind.
- Semiconductor margins are a single-variable risk. Each 1 pp change in semiconductor gross margins shifts S&P 500 EPS growth by roughly 1 pp. A normalization from today's 70% back to the 15-year average of 55% would cut S&P 500 earnings by about 10%.
- "Other income" will create a misleading year-over-year comparison in Q2 2027. The $150 billion in unrealized private equity gains that lifted Q2 2026 EPS will not recur at the same scale. Year-over-year comparisons will look sharply worse in mid-2027, even if underlying business performance holds.
- Valuations price in skepticism, not complacency. The S&P 500 trades at a P/E discount to what a macro model implies given today's record return on equity of 24%. The market appears to be discounting some mean reversion in profitability. That embedded skepticism is part of why Goldman's 12-month return forecast hinges on earnings growth, not re-rating.
Footnote:
1 Snider, Ben, Ryan Hammond, Daniel Chavez, Kartik Jayachandran, and Christophe Sung. "Over-Earning, but Not an Earnings Bubble." Goldman Sachs Portfolio Strategy Research, 17 Sept. 2026, https://www.gspublishing.com/content/research/en/reports/2026/09/17/3eb7e544-9a8d-4474-81a2-16288ad509c6.pdf.
