AI’s Trillion Dollar Data Center Gamble Risks Debt, Bankruptcy and Stranded Assets
AI hyperscalers may spend $5 trillion in four years, but weak revenue, rising debt and obsolete chips put companies, ratepayers and the US economy at risk.
Summary
Wharton finance professor Jessica Wachter and a collaborator estimate hyperscaler spending will approach $1.1 trillion through 2027. Covering capital costs, depreciation and a 15% return requires Alphabet, Microsoft, Amazon, Meta and Oracle, OpenAI’s partner, to lift their productivity 2.7 times by 2030, requiring a decade of US IT boom growth within several years. Failure risks missed interest payments, bankruptcy and history’s largest capital misallocation. Spending is about $750 billion in 2026, against $150 billion to $200 billion of AI revenue. Four year projections exceed $5 trillion, near 3% of US GDP.
Stijn Van Nieuwerburgh’s scenario of 183 gigawatts built from 2025 to 2032 at $41 billion per gigawatt requires about $3.7 trillion in annual revenue by 2032, assuming a 10% return. GPUs represent about 60% of data center costs and performance doubles every two years, forcing repeated upgrades or risking stranded hulks, Princeton’s Mihir Kshirsagar warns. Morgan Stanley calculates external capital will fund more than half of $2.9 trillion spent from 2025 to 2028. Alphabet’s nearly $120 billion quarterly revenue still produced a $5.9 billion free cash deficit, its first since Google went public in 2004.
A survey of about 6,000 executives in the US, UK, Germany and Australia found around 90% saw no productivity gain over three years, but they expect 1.45% over the next three, including 2.25% in the US, alongside $280 billion of private AI spending by end 2026, higher sales and substantial job cuts. Meta’s Hyperion project in Richland, Louisiana, grew from $10 billion for two gigawatts in late 2024 to $50 billion for five in July. Its Blue Owl venture and four year leases spread financing risk, while Entergy plans ten gas plants totaling 7.5 gigawatts despite concerns over residential bills. Gary Gensler expects eventual retrenchment. Cheaper models, weak demand and public backlash could strand assets.
Positives
- $280 billion in private sector AI spending is anticipated by the end of 2026 as businesses increase investment.
- 1.45% productivity growth is expected by surveyed executives over three years, rising to 2.25% among US respondents.
- GPU performance roughly doubles every two years, supporting continued improvements in AI model capabilities.
- Meta’s 20 year electricity purchase guarantee is intended to cover Entergy’s power plants and related infrastructure.
- Alphabet, Microsoft, Amazon, Meta and Oracle retain substantial earnings and deep financial resources, limiting immediate balance sheet pressure.
Risks & concerns
- $750 billion of 2026 infrastructure spending vastly exceeds estimated annual AI revenue of $150 billion to $200 billion.
- 2.7 times productivity growth by 2030 is required for hyperscalers to cover capital, depreciation and a 15% return.
- More than half of $2.9 trillion in 2025 to 2028 spending may use external capital, distributing risk through lenders, funds and insurers.
- Alphabet recorded a $5.9 billion free cash deficit despite nearly $120 billion in quarterly revenue, its first shortfall since 2004.
- GPUs account for about 60% of data center costs and rapidly lose competitiveness as performance doubles approximately every two years.
- Meta’s $50 billion Hyperion expansion and Entergy’s ten planned gas plants could expose Louisiana residential customers to surplus power costs.