AI data center boom sparks fears of glut amid lending frenzy

A rapid, deeply financed build-out of AI data centers is under way, with industry estimates of infrastructure needs ranging from roughly $5–10 trillion (Brookfield) to about $7 trillion by 2030 (McKinsey) and outsized spending plans from major players, prompting a flood of debt and structured financings. Market participants warn of classic boom risks — overbuilding that could render assets uneconomic, high leverage, refinancing risk in three to five years, opaque tranching and synthetic leases that push liabilities off balance sheets, and even borrowers seeking more than 100% of build costs — while tight yields (sometimes ~100bp over Treasuries) and stretched deal structures raise underwriting concerns. Regulators are taking notice, and investors face potential significant downside if demand or technology underperforms and lenders retrench, creating ripple effects across credit and equity holders.

Analysis

A rapid, deeply financed build-out of AI data centers is underway with industry estimates ranging from $5–$10 trillion (Brookfield) to roughly $7 trillion by 2030 (McKinsey), and outsized spending plans such as OpenAI’s cited $1.4 trillion. Lenders have placed at least $175 billion of U.S. data-center credit this year, and market participants report some financing yields only ~100 basis points over Treasuries, indicating strong appetite despite long-duration technology risk.

Market participants are increasingly using synthetic leases, securitizations and tranche sales to move liabilities off corporate balance sheets and distribute risk to insurers and pension funds, while some borrowers are seeking >100% of build costs (one request reportedly for 150%). Executives including Howard Marks and Sadek Wahba warn that deal nuance and underwriting discipline matter amid swapped and opaque structures such as master trusts and rotating asset vehicles.

Key risks are a potential glut from overbuilding that could render assets uneconomic, concentrated refinancing needs in three-to-five years if lenders retrench, and growing regulatory scrutiny (Bank of England review) and market jitters after spending-led hits to Meta and Oracle. These dynamics raise medium-term credit and equity downside if demand or technology underperforms or if transparency and underwriting weaken.

Key Decisions for Investors

• Reduce or hedge exposure to long-duration, low-spread data-center debt and favor shorter-duration or higher-spread senior secured paper where underwriting is transparent
• Avoid complex securitized tranches or master-trust structures without full cash-flow stress tests and covenants, and demand documentation that prevents rapid asset rotation
• Be cautious on early-stage or high-valuation equity plays tied to unbuilt facilities (e.g., recent Fermi-type listings) and monitor earnings/capex signals from hyperscalers like Meta, Oracle, Microsoft and Google for demand confirmation
• Monitor refinancing calendars over the next three to five years and regulatory actions (Bank of England reviews, yield spread moves); use these triggers to scale risk on or off and to reprice exposure