Over the weekend, we published a length report analyzing what we dubbed "the $1.8 trillion off-balance sheet time bomb at the heart of the AI supercycle",  which focused on the recent surge in popularity of off-balance sheet SPVs, as well as $1 trillion in long-term purchase commitments, and $800bn in lease commitments, which support the AI buildout, and do so by masking the true funding costs of the AI buildout, yet which add significantly to off-BS operating leverage. Additionally, when one adds billions in variable lease payments, over $100BN in embedded capex inside accounts payable, as well as the Construction-in-Progress pipeline which reflect capacity built but not yet expensed - and the $520bn cumulative depreciation estimate is where it eventually lands - and one gets a far more ominous picture of the capital stack that is backstopping the biggest infrastructure buildout in US history, which according to Goldman will translate into as much as $1.4 trillion in 2027 capex, and much more in the following years.

We summarized the complex funding picture which is effectively a series of circular financings across the entire AI ecosystem, "vendor financing and repurchase-style arrangements mean a single counterparty's stress can propagate through several balance sheets at once. Think of it as rehypothecated leverage. Concentration compounds it: the >$2tn of remaining performance obligations across major AI players is built on a handful of very large, long-duration contracts, so the backlog that justifies the spending is also a concentrated counterparty exposure."

But even without focusing on the potentially dire consequences of this off-balance sheet funding unwinding, there is a more mundane risk propagating from the AI buildout, and it has to do with the massive leverage unleashed in public debt markets. We spent the first part of our post discussing just that, but the punchline was captured by the following three charts, the first of which shows that YTD, some $236BN in AI-linked debt has been issued, a 357% increase from the same period last year. By year-end, MS expect this number to more than double to $570 billion.

The second, and perhaps most important chart, is the one showing the dramatic increase in hyperscaler gross leverage, which has surged from 0.9x in Q3 '25 to 1.8x currently, doubling in just over two quarters, and surpassing the gross leverage of the entire energy sector. At this rate, hyperscaler debt is growing at about 0.3x turn per quarter.

Not surprisingly, as a result of the surge in combined leverage, hyperscalers are drifting wider, and after trading inside AA spreads for much of 2025, are now on top of A, and as MS warns, "may widen further on supply." And it's not just outlier Oracle: META is now trading wider to CDX IG. 

Putting it all together, what the above indicates is that hyperscalers, and the broader AI ecosystem, has gong into an all-out debt expansion mode, using both on and off-balance sheet vehicles, to fund as much of the infrastructure buildout (now that the price of 1GW of data center has increased from $50BN most recently to as much as $100BN, due to the jump in silicon density with NVDA Rubin-class racks approaching 600kW and every gigawatt carrying far more GPU and HBM content) while the window is open, and - well - while they can.

Today Nvidia could, and according to Bloomberg, Nvidia became the latest company seeking to raise at least $20 billion from its first corporate bond sale since 2021. 

The chipmaker is marketing bonds in seven tranches, with maturities spanning two to 30 years, Bloomberg reports. Price talk on the longest tenor of the Aa1/AA-rated company is a spread of about 0.9% above Treasuries, while the FT adds that "the 10-year portion of the bond was expected to yield 0.75 percentage points above US Treasuries during initial price discussions."

As Bloomberg writes, picking up where we left off, "the deal is extending a relentless wave of borrowing for companies spearheading the artificial intelligence boom. Firms such as Alphabet Inc. and Amazon.com Inc. have been tapping every corner of the debt market to build the computing capacity needed for AI’s rapid expansion, having raised hundreds of billions of dollars since last year. Investors have readily absorbed the supply."

Nvidia said that it intends "to use the net proceeds from this offering for general corporate purposes, including repayment and refinancing of outstanding notes."

Nvidia last tapped the investment-grade bond market in June 2021, when it raised $5 billion. Its ambitions now are far greater.

And yet, as the FT cautions echoing our own warning, "early signs of market fatigue have prompted some tech companies to find alternative avenues for financing."

Anthropic has turned to private credit investors to seal a $35bn deal backed by Broadcom. Google’s parent Alphabet decided to issue equity for the first time in more than two decades, bringing in $85bn in fresh capital earlier this month.

In any case, the bond market clearly remains open for now; however with leverage within the AI space exploding at the fastest pace on record, it may not take much for the window to close at which point the question will again re-emerge: how will the handful of AI-leading companies fund the trillions in unfunded future obligations.