Overview

• Citi now projects more than $2.8 trillion in AI-related infrastructure spend through 2029 and estimates hyperscalers will invest about $490 billion next year, up from a prior $420 billion view.
• The bank cites mounting enterprise adoption as a driver for bigger outlays, even as companies guide investors to spending that runs ahead of visible demand.
• Citi estimates roughly 55 gigawatts of additional power capacity will be needed by 2030, with costs near $50 billion per gigawatt, a burden already pressuring free cash flows.
• Debt-led funding is accelerating, exemplified by Oracle’s $18 billion bond sale to expand cloud capacity tied to major AI commitments.
• Unconventional structures are proliferating, including a reported OpenAI–Nvidia arrangement to deploy 10 GW over five years alongside a $100 billion equity investment and potential chip leasing, while Bain flags a large revenue shortfall by 2030 that has revived bubble comparisons.