Build it and they
will prompt.
This year, five American companies are on course to spend nearly as much building AI infrastructure in one year as the United States spent building the entire US Interstate Highway System, a project that spanned 36 years.
One year of estimated AI capex vs. US megaprojects.
Five companies. One year.
On method: Each figure estimates the AI / datacenter-attributable portion of company-wide capital spending. The April allocation framework is retained: Amazon ~60%, Alphabet ~85%, Microsoft ~90%, Meta ~95% and Oracle ~95%. Those shares are applied to refreshed guidance or actuals; ranges use their midpoint. Aggregate: ~$625B AI-attributable on ~$761B total capex. Companies do not disclose a comparable AI-only subtotal, so this is a transparent estimate, not a reported accounting figure. Attribution support comes from company spending-mix statements: Amazon identifies AWS plus non-AWS investment; Alphabet says the vast majority is technical infrastructure; Microsoft identifies GPUs, CPUs and long-lived datacenter assets; Meta cites component and datacenter costs; Oracle ties its capital program to AI cloud infrastructure. Reporting calendars differ, making the comparison directional.