AI titans’ ‘circular deals’ are starting to look like ‘daisy chains’

The "circular deals" among AI industry leaders are beginning to resemble the "daisy chains" that sparked the 1980s savings-and-loan crisis. Back then, interconnected transactions masked risks, amplified systemic dangers and fueled asset value growth before nearly a third of US thrift banks failed. Today, a web of multi-year commitments involving financing, semiconductor purchases, power procurement and data center leases creates a similar interconnected ecosystem.

OpenAI has struck arrangements with Nvidia, CoreWeave and others prior to its planned IPO next year, while competitors Anthropic and xAI have pursued comparable agreements. Anthropic signed a $35 billion cloud-computing deal with Lambda, a Nvidia-supported neocloud, in recent months. Nvidia, involved in over $750 billion worth of AI investments, financing deals and partnerships, accounts for roughly a quarter of its business next year, CFO Colette Kress said.

The AI investment spurt is unprecedented in scale, quadrupling compared to three years ago, according to Bank for International Settlements data. This raises concerns: how much of the industry's growth is fueled by the same entities that benefit from it? Hidden risks are intertwining with a borrowing spree. The five largest hyperscalers owe roughly $1.65 trillion through special purpose vehicles and other off-balance-sheet structures, surpassing their directly reported $1.35 trillion debt, according to Nikkei.

Circular transactions can boost valuations and revenue expectations, justifying more investment. Financing creates demand, demand supports valuations, and valuations enable further borrowing, which fuels more demand. While not fraudulent "round-trip" transactions, they share similarities with the deceptive accounting of the 1980s S&Ls, which traded junk bonds within a reinforcing "chain", creating an illusion of liquidity, solvency and market stability.

However, hyperscalers are solvent, generating unprecedented revenue growth amidst global demand for AI. Interest rates have not yet played a role, and no regulator has accused them of fraud. Wall Street offers multiple financing tools. Despite these differences, circular deals share the dangerous dynamics of the thrift-era daisy chains.

They mutually validate asset values with correlated counterparties, potentially obscuring where risk ultimately lies and concentrating risks unevenly across the chain. If one critical link fails – such as a hyperscaler losing demand for its products or services – the flywheel can turn vicious. The interconnectedness that fueled growth can also transmit distress throughout the ecosystem, turning one company's losses into another's impaired investments, contract defaults or stranded assets.

Nvidia President and CEO Jensen Huang insists these deals are not circular as they operate independently, but the high demand for compute power and Nvidia's substantial backorders through 2027 suggest the risk of a vicious cycle looms. The AI boom could generate sufficient revenues to make today's commitments seem modest in hindsight, but assuming consistent demand, falling computing costs, and productive infrastructure is precarious.

The 1980s thrift crisis lesson isn't that AI's virtual flywheels are doomed to collapse, but that interconnectedness can turn an unexpected setback into a far larger problem. Breakthroughs requiring less computing power, AI pricing collapses, or unforeseen shocks could abruptly reverse today's economic trajectory.