US Hyperscalers' AI Spend Strains as DRAM Demand Soars

Jefferies reports signs of financial strain among US hyperscalers as they absorb nearly all their operating cash flow in capital expenditures for AI infrastructure. By 2026, projections indicate hyperscalers will spend 92% of their operating cash flow on capex, up from 41% in 2023. Total annual capex is expected to reach $700 billion this year and $800 billion next year, equating to 2% of the US GDP. The surge in spending is directed toward memory chips, particularly DRAM, with hyperscalers potentially allocating 28% of their cash flow to it. Supply constraints in the DRAM market, exacerbated by the slowing of Moore's Law, prompt companies like Nvidia to secure long-term supply deals amid fears of overinvestment and uncertainties about AI monetization.