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

Robin Wigglesworth reports that the combined off-balance sheet lease obligations and purchase commitments of hyperscalers rose from roughly one trillion to roughly 1.5 trillion dollars between the first and second quarter, based on Goldman Sachs figures compiled from regulatory filings. Of the 1.5 trillion in lease obligations, approximately 500 billion covers leases already started and appearing in financial accounts, while approximately one trillion covers leases not yet started and disclosed only in footnotes. Purchase commitments for chips, equipment, cooling, and power also rose to roughly 1.5 trillion, with Google alone accounting for approximately 800 billion of that total. The disclosure is not hidden but is inconsistent across companies, with Google being the most transparent and others disclosing only that material payment obligations exist without specifics.

The lease structures keep obligations off the balance sheet so they do not appear as debt, but purchase commitments often cannot be easily exited and therefore, in Wigglesworth's words, talk and quack like debt without appearing as debt. Meta's data center in Louisiana called Hyperion illustrates the structure: Meta invests only 20 percent of the cost but guarantees a 20-year lease covering the full cost through a joint venture with Blue Owl. Hyperscalers have already surpassed last year's record for bond sales and Wigglesworth says the scale of bond issuance is showing signs of indigestion, driven by the fact that data center investment now exceeds what free cash flow alone can fund. He cites a Claude estimate that roughly 80 percent of the 500 billion AI investment cycle is expected to be debt and only 20 percent equity, and argues that debt-fueled capex cycles are more dangerous than equity-financed ones even when the underlying technology proves transformative, contrasting this with the dot-com bust, which caused a 50 percent stock market drop but was economically a near-nothing event because it was equity financed.

On compute as collateral, Wigglesworth says commercial real estate and pipeline infrastructure are understood collateral whereas compute depreciation and maintenance risk for lending purposes is less proven. He notes that every data point over the past nine months has not supported the Michael Burry depreciation thesis on AI chips, pointing to CoreWeave leasing six-year-old chips through 2029 at favorable rates as a contradicting example, but cautions that Burry being wrong so far does not mean a supply glut cannot occur or that lending a trillion dollars against compute is sound. He describes CoreWeave as having the biggest gap between EBITDA and net income loss he has ever seen and characterizes its debt profile as railway-like in its extremity, while distinguishing it from Facebook, Alphabet, and Amazon, which he says are not at risk of going bust because existing products like YouTube provide a meaningful fallback.

On private credit, Wigglesworth says the asset class is a genuinely good idea that de-risks the financial system by moving bundles of risk into the non-bank ecosystem, but that too much money flooded in too quickly and was in some cases deployed poorly. He says private credit's apparent low volatility is an artifice created by the absence of mark-to-market accounting and that a default cycle will likely be far worse than backward-looking historical numbers suggest. Over half the high yield bond market is now rated double-B and has become far more solid than it has ever been, while private credit has absorbed the dicey lending that used to sit in high yield. Origination in private credit shifted toward spray-and-pray because lenders earned fees by sourcing borrowers regardless of quality, a dynamic he compares to 2008 mortgage origination. He says money flooding into private credit in 2022 refinanced 2018 loans and kept defaults artificially close to zero, and that defaults will likely be higher as inflows decline. He predicts a bad default cycle has started but is being masked and will not be catastrophic, and that public BDCs could trade at 30 to 40 cents of net asset value in a crisis, creating opportunity for investors with the stomach for it. He identifies a specific concern in private label credit ratings, where insurance companies internally label direct loans as investment grade, and notes that many private insurance companies are owned by private equity firms that sometimes also own the private label ratings companies, creating a tangled conflict of interest.

Wigglesworth draws historical parallels to illustrate how debt cycles around transformative technologies have repeatedly ended in crisis without negating the underlying technology's value. US railways in the 1870s and 1890s issued the equivalent of around 10 trillion dollars of bonds in today's terms, and the 1871 crash caused the collapse of Jay Cooke, which he describes as equivalent to J.P. Morgan going bankrupt overnight. The 1870s downturn was originally called the Great Depression before being renamed the Long Depression after the 1930s event. Despite the losses, the physical railways remained and transformed the US economy. Wigglesworth argues that the optimal number of financial crises is arguably not zero, as they can drive transformative economic development, though guaranteeing zero crises would require banning speculative activity, which carries its own significant downsides.

His book, A Fabulous Debt, traces the history of bonds from their origins in Renaissance Italy through to the present day, arguing that the bond market has supplanted the banking system as the dominant credit engine of the global economy and that bonds are the original decentralized finance.