Can markets find the AI cash?
Richard Waters frames the next leg of US AI infrastructure spending as a financing test, as hyperscalers prepare to raise spending again in 2027. In the latest quarter, Alphabet, Amazon, Meta, Microsoft and Oracle committed $33bn more to capex than they generated from operations, the first time their combined free cash flow turned negative. Meta's $95bn stock issuance is cited as a sign of new reliance on outside finance. The leading AI spenders are expected to invest more than $800bn in 2026. One forecast puts the build-out at about $5tn by end-2030, $4tn of it debt, which Theory Ventures' Tomasz Tunguz compares with the entire US municipal bond market. Columbia's Stijn Van Nieuwerburgh puts facilities already in planning for 2026–2032 at an average 3.6% of GDP a year. The bear case: higher rates, circular deals and off-balance-sheet commitments; Alphabet lost $500bn of market value in one day after an executive said AI search generated less advertising than traditional search. The bull case: fast AI/cloud revenue growth and high gross margins (Nvidia 75%, Anthropic 85%, SpaceX's AI division 65%, with a reported payback of under a year on its data centres). Voter-backed local moratoria (more than 40 in the midterms) could delay revenue but also keep supply tight and margins high. Caveats: this is opinion; the $5tn/$4tn totals are forecasts, not committed financing, and the piece sets out competing scenarios rather than a settled funding gap.
Why it matters
The first quarter of negative combined free cash flow at the five biggest spenders means the AI build-out now depends on capital markets, not just hyperscaler cash. That matters for everyone downstream (neoclouds, data-centre developers, chip and power start-ups) whose growth relies on that capex continuing. It complements yesterday's bank-insurance and CoreWeave debt rows with the top-down numbers.