Big Tech's Capex Is Half of Wall Street's Profit Growth

Goldman Sachs projects hyperscaler capital expenditures rising from roughly $800 billion today to $1.2 trillion next year and $1.4 trillion by 2028, and finds that a large share of recent S&P 500 earnings growth is tied to that spending. The mechanics matter: suppliers book full revenue immediately when they sell chips, servers and power gear, while hyperscalers capitalize and depreciate those costs over years. That timing gap inflated S&P 500 EPS - which jumped unusually (51% YoY in Q2) - but as capex growth rates slow, depreciation from prior years will subtract from earnings. Goldman estimates next year’s capex adds about 11 percentage points to S&P 500 EPS growth while hyperscaler depreciation removes roughly 5 points, and by 2028 AI investment could become a slight net drag.

Two other forces also temporarily boost earnings: roughly $150 billion of unrealized gains on private-equity stakes (about 12% of S&P EPS) and semiconductor margin expansion (memory gross margins near 80%) that may erode as capacity ramps. Goldman judges hyperscalers need ~ $300 billion of annual AI revenue to break even on infrastructure, while truly adequate returns and healthy downstream margins would require roughly $1 trillion a year in AI application spending - about two-thirds of current global software spend. The crucial next signals are capex guidance alongside actual cloud and AI revenue growth and where end users reallocate budgets.