Robin Wigglesworth on Hyperscalers' 1.5 Trillion of Off-Balance Sheet Liabilities, Private Credit, and His Book "A Fabulous Debt"

Hyperscalers’ data-center buildout is turning into a major debt-financing cycle, Robin Wigglesworth says. Companies that once funded investment from free cash flow are increasingly tapping bond markets: their bond sales totaled several hundred billion dollars last year and had already surpassed the previous annual record this year. Wigglesworth sees the shift as evidence that these businesses are becoming more like capital-intensive utilities, even though returns on some data centers are currently strong. He warns that a successful technology can still produce a damaging debt cycle.

He estimates hyperscalers’ lease-related obligations at about $1.5 trillion: roughly $500 billion for leases already started, and about $1 trillion for leases not yet started, disclosed in footnotes. Separately, purchase commitments for chips, equipment, cooling and power have grown from about $1 trillion to $1.5 trillion. Such commitments can act like debt, he argues, because they create difficult-to-avoid payment obligations despite not being reported as debt. Google discloses about $800 billion of these commitments, nearly half the total, including about $200 billion labeled short term; Wigglesworth believes that likely means payments due within roughly 12 months. Varying disclosure language and detail make comparisons difficult.

Meta’s Hyperion data center in Louisiana illustrates the financing structures involved: Meta invests about 20% and guarantees a 20-year lease in a project associated with a Blue Owl joint venture. Such arrangements can transfer the support of highly rated hyperscalers to lower-rated developers or neoclouds, while developers’ reported backlogs may depend on hyperscaler commitments. Wigglesworth says these structures are not necessarily improper or new in kind, but their scale is unprecedented. He would prefer clearer disclosure and more plain-vanilla public bond financing; a mainstream company would generally pay less to borrow $10 billion through a corporate bond than from a handful of private-credit firms, though private financing can be faster and more flexible.

The risks differ across companies and assets. Wigglesworth is relatively unconcerned about Facebook, Alphabet or Amazon if AI investment disappoints, given their established businesses, but sees more vulnerability in indebted outliers such as CoreWeave, which he says may lack fallback businesses and has an unusually large gap between EBITDA and net income loss. He regards Oracle as the weakest of the hyperscalers financially in debt relative to revenue, while stressing that it is not a small or bad company. Lending against compute is harder to assess than lending against commercial property or oil infrastructure because chip lifetimes, maintenance and replacement are less straightforward. Six-year-old chips being leased through 2029 at favorable rates have not supported Michael Burry’s depreciation argument, but present demand does not rule out a future glut or prove that trillion-dollar lending against compute is sound. Wigglesworth has no clear view on when an AI capex correction might arrive; much depends on how transformative AI proves to be.

Private credit can move lending risk out of banks and reduce risk in the financial system, Wigglesworth says, but too much capital arrived too quickly, and some lending was poorly deployed as firms chased fees. He expects defaults to be materially worse than backward-looking figures suggest, partly because assets are not consistently marked to market and fresh inflows have helped refinance older loans. He considers many assumed recovery rates unrealistic, especially for software companies without tangible assets. High-yield bond quality has improved, with more than half the market rated double-B, while private credit has taken on riskier lending. He opposes selling retail investors semi-liquid vehicles holding five-year loans: the investment should lock up money for five years. Liquidity mismatches and leverage are vulnerabilities, particularly when leveraged funds lend to already leveraged companies. He expects a bad default cycle that has started but is being masked, though not a catastrophic one; public BDCs could fall to 30–40 cents of net asset value in a crisis, and distressed prices may keep falling.

In A Fabulous Debt: The Epic Story of How Bonds Built the Modern World, Wigglesworth argues that bonds are a powerful, decentralized way to pool savings and have become the global economy’s dominant credit engine, but their reputation for safety can foster false confidence. The Erie Canal’s success encouraged US states to borrow for canals, banks and industrialization, before widespread defaults: roughly half the states defaulted in the 1840s, and some never repaid. Railway booms also brought losses and bankruptcies even as railways transformed the economy; he estimates bonds issued in the 1870s and 1890s would equal about $10 trillion at today’s US-economy scale. He sees financial engineering as a continuing effort to optimize risk and reward or game rules, but says major crashes such as 2008 are rare. Private-label ratings on private-credit loans concern him when they can make loans appear investment grade and reduce modeled capital or default-risk charges, especially amid links among private-equity owners, lenders, rating providers and insurers.