Q3 S&P 500 EPS Is Tracking 30% — But Half of That Came From AI Capex You'll Keep Paying For

Goldman Sachs Research published a Sept 23 analysis making the cushion problem explicit. They estimate that \"almost half of S&P 500 growth in EPS in 2026 comes from AI investment\" — meaning the 30% expected Q3 print is being lifted by the same revenue lines that are being funded by the same capex lines that everyone else is watching for downside. Goldman further sizes the sensitivity: every $250 billion surprise in AI capex moves S&P 500 earnings growth by roughly 6 percentage points, almost exactly the spread between a ho-hum and a blowout quarter.

The forward path matters more than the Q3 number. Goldman projects hyperscaler capex at $1.2 trillion in 2027 and $1.4 trillion in 2028, with the AI investment contribution moving from \"a tailwind this year to a marginal drag\" by 2028. Their S&P 500 EPS path rises to $415 in 2027 and $460 in 2028 — growth of 13% and 11%, well below the 29.5% setup in front of investors today.

The Q2 2026 cash flow statements tell the story one beat before Q3 prints. The big four (Alphabet, Amazon, Meta, Microsoft) together spent $165.0 billion on property, plant and equipment in the June quarter — a 87% jump from a year ago, according to each company's 10-Q filings. Annualized, that puts their combined run rate above $660 billion and reinforces the $720–$760 billion consensus tracked by TMT Finance and Statista for the full 2026 calendar year — versus Goldman's $800B figure that also captures Oracle.

Operating leverage is the cushion under the 30% EPS story — and it is the cushion that is starting to give. Meta Platforms still produced $1.7 billion of free cash flow in Q2 2026, but that was down from $9.0 billion a year earlier, an 81% collapse. Alphabet's $44.9 billion capex bill pushed FCF negative for the first time in recent memory at -$5.9 billion versus a positive $5.3 billion last year. Amazon printed -$8.8 billion in Q2 free cash flow versus essentially breakeven a year ago. Only Microsoft ($19.6 billion) still generates substantial FCF, and even Microsoft's was down 24% year-over-year in its Q4 FY2026 reporting period.

Q3 2026 is the first reporting cycle where D&A, interest expense, and stock-based compensation from the latest $165B capex bill start to flow through. The reported earnings power will still look strong — JPMorgan is on track for $5.94 EPS, up 17% year over year on revenue growth near 9%; NVIDIA already printed Q2 FY2027 revenue of $96.2B, up 106% from a year earlier. The cushion between those numbers and 2027 estimates, though, depends on hyperscaler revenue growing fast enough to absorb both the depreciating asset base and the new debt funding the cycle.

The narrow-rally thesis is the second leg of the cushion argument. Reuters reported on Oct 12 that more than 200 S&P 500 companies — about four in ten — are still in the red for 2026 even with the index up 14%, because the technology sector alone delivered more than 10 percentage points of that return. A 30% EPS print built largely on hyperscaler capex into a market where multiple expansion has gone the wrong direction creates a fragile handoff into 2027.

The composition of the 30% matters as much as the headline. Information Technology, Communication Services, and Energy are the three sectors driving the bulk of the growth, with Energy alone expected at +114.7%. If hyperscaler AI commentary underwhelms, the 30.6% headline drops quickly toward the 24% zone. If it surprises to the upside, the cushion stretches — but only into 2027, where Goldman's $1.2 trillion capex forecast sets a much higher bar.