A residual-value guarantee does not sit in the debt column, which is roughly why it can be three times the size of the debt and draw a fraction of the attention [18]: the chip revenue is recognised on delivery and the risk that the hardware is worth less than promised later stays behind, so the $105bn attached to one OpenAI initiative [6] is a short position on the resale value of Nvidia's own product, written by the party with the strongest commercial reason to insist that resale value holds.
The borrowing is the legible part. Gross debt of $33.5bn against $8.5bn at the prior fiscal year end [4][5] is a $25bn increase, which is the June 2026 senior unsecured note issue to the dollar [3][1], so the balance sheet expansion is that one transaction and the book is 3.9 times its prior size [2]. Put the guarantee next to it and 105 over 33.5 is 3.1 [3]. Put the top of the $119bn to $279bn multi-year service commitment range [7] next to the quarter that just printed [1] and you get 2.9 quarters of revenue [4]. The $80bn repurchase authorization the board approved in May 2026 [8] is 2.4 times gross debt [5], and it lands exactly on the low end of the $80bn to $200bn peak liability estimate [2].
The share price registered none of this with much conviction, since $196 to $219 in late July and early August [9] is an 11.7% band [6] against an estimate range that spans 2.5 times [7]. Direction is cheap to price, but the magnitude here can only be judged once the guarantee's terms are known. And the report's description of the OpenAI side, equity investments of roughly $30bn plus conditional guarantees tied to project financing [13], is ambiguous about which direction the equity travels, which matters a great deal, because if Nvidia is funding the customer whose project is backstopped by Nvidia's guarantee, the two exposures are one exposure counted twice.
This is probably wrong, but the guidance is the strongest argument against the credit read rather than for it: $108bn plus or minus 2% for the third quarter [15] is a 12.3% sequential step [8], and approximately 70% growth projected for fiscal 2028 [16] would turn even $200bn of peak liability into a few quarters of revenue. The counter-thesis is the one Cramer supplied himself before telling holders to stay long, comparing the structures to the late-1990s vendor lending at Cisco, Lucent and Nortel, all of which suffered when the cycle turned [10][11][12]. Data centre revenue of $89bn is 92.5% of the total [14][10], so the revenue concentration and the guarantee concentration are pointed at the same short list of counterparties, or rather at the same short list of funding rounds.
What would falsify the credit story is narrow and checkable: guarantee terms with disclosed triggers showing the exposure only bites below some utilisation floor, and off-balance-sheet commitments growing more slowly than the operating cash flow meant to stand behind them [17]. Those terms are not in the record here, which leaves the record with a 3.1 [3].
