US hyperscaler companies are projected to spend $800 billion on capital expenditures in 2026, a 94% increase from 2025, driving roughly half of S&P 500 earnings growth through chipmakers and tech suppliers. However, Goldman Sachs forecasts slower capex growth in subsequent years, which combined with rising depreciation charges will dampen AI investment's contribution to earnings. Market analysts warn of structural headwinds including valuation compression from higher interest rates and massive new equity supply from IPOs and insider lockup expirations.
JPM analyzes sustainability of hyperscaler AI capital expenditure and memory demand, presenting bullish and bearish scenarios across five key debates including AI capex viability, memory budget allocation, HBM specifications, long-term agreements, and China competition. JPM leans bullish on memory demand through FY28 with robust hyperscaler investment, though expects moderating ASP growth and views China as a structural but near-term limited threat due to technology gaps.
Two X posts debate hyperscaler AI capital expenditure trends. One challenges claims that 50% of AI chips sit idle in warehouses, arguing the lag between purchase and deployment reflects normal setup timelines. Another projects memory will surge from 14% of AI capex in 2025 to 64% by 2027, driven by GPU/TPU demands for stacked HBM and creating supply shortages through the decade's end.
UBS forecasts AI infrastructure investment (capex) will grow 84% this year and 33% next, while Barclays warns the AI trade is entering a mature phase with execution risks. The Federal Reserve's first rate increase since 2023 is seen as a credibility-clearing event that supports the dollar and equities, though concerns persist about whether hyperscalers can generate sufficient revenue to justify massive capex spending.