Goldman Sachs has tracked $1.5 trillion in lease commitments tied to AI hyperscalers' data centre build-out, with $1 trillion of that sum absent from the companies' financial statements under current accounting rules. Morgan Stanley separately counted $982 billion in purchase commitments for computing power, chips and equipment across five hyperscalers.
Goldman Sachs analysts have tallied $1.5 trillion in aggregate lease commitments tied to the data centre build-out by AI hyperscalers. Of that total, $1 trillion has not yet commenced and therefore does not appear in the companies' financial statements, even though the payments will still come due.
Off-balance-sheet by design
Under GAAP ASC 842, a lease obligation only lands on a balance sheet once it actually starts, not when the agreement is signed. Legally binding but not-yet-started leases sit in the footnotes instead, which is where Goldman found them. According to Goldman Sachs: "From a credit perspective, this treatment can understate leverage and future liquidity needs", the firm's analysts wrote.
Fitch and Moody's also exclude these leases from their metrics until the leases start, while S&P applies a more conservative standard, counting a lease when it judges the obligation effectively equivalent to debt. Total lease commitments have grown from about $200 billion five years ago to today's $1.5 trillion. Goldman estimated $1.2 trillion just last month, of which $750 billion had not yet commenced.
The Beignet blueprint
Meta set the template with its Hyperion data centre in Louisiana, financed through a joint venture with Blue Owl called Beignet. Meta owns just 20% of Beignet but committed to lease Hyperion for at least 20 years, a guarantee that let Beignet issue a $27 billion amortising bond that never shows up as debt on Meta's own balance sheet. Other hyperscalers have since explored their own increasingly creative financing, with lease structures only loosely modeled on Beignet.
Purchase commitments add up
Beyond leases, hyperscalers have also promised to buy computing power, chips, equipment and electricity. Morgan Stanley's credit analysts counted $982 billion of these purchase commitments across Alphabet, Microsoft, Amazon, Nvidia and Oracle at the end of the first quarter. None of it counts as conventional debt on a balance sheet, yet all of it will eventually need paying.
Hyperscalers still carry an average net leverage ratio of just 0.5 times, against 0.8 times for the technology sector and 1.8 times for US non-financial companies generally, according to Morgan Stanley. Even so, the pioneering Beignet bond is trading at a yield of 6.95%, up from a low of 5.65% when it was issued last autumn.
Source: Financial Times
Trading involves risk.