Nvidia's spotless balance sheet has about $30bn parked just outside it

Investors have started reading Nvidia's footnotes instead of its balance sheet. According to Crypto Briefing, the company's purchase obligations reached nearly $30 billion, mostly long-term supply contracts, capacity reservations and multi-year cloud agreements that are contractual cash outflows but not reported debt ^[1][2].

The composition matters more than the headline number. The report puts $18.8 billion in inventory purchase commitments and $8.8 billion in multi-year cloud service agreements ^[3][4], which leaves roughly $2.4 billion of the total unallocated in the account given ^[5]. These are promises to spend whether or not GPU demand holds ^[2].

The accounting consequence is where the argument gets sharp. Crypto Briefing calculates that folding those obligations into the picture would cut reported shareholder equity from $49.1 billion to about $19.8 billion, a drop of roughly 60% that would mechanically inflate price-to-book ^[6][7]. That is a presentation adjustment rather than a cash event, but it is the sort of adjustment that reprices a stock held for its capital efficiency.

There is already evidence the commitments bite. The same account reports $4.5 billion of charges for excess inventory tied to purchase obligations in the first quarter of fiscal 2026 ^[8], equal to about 24% of the $18.8 billion inventory commitment line ^[9]. Worth noting that the report dates the obligation total to April 28, 2024 while the charge sits in fiscal 2026, so the two figures are not from the same reporting period and the current balance may differ ^[10].

The second layer is the financing platform. Nvidia has announced a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR intended to mobilise more than $500 billion for AI compute infrastructure by August 2026 ^[11]. The structure is designed so the debt sits with the buyers: data centre operators and cloud providers borrow from the financial firms to purchase Nvidia hardware, with compute capacity treated as a financeable asset class ^[12]. That target is more than sixteen times the disclosed purchase obligations, so the off-balance-sheet exposure that matters is not Nvidia's own commitments but its customers' ^[13].

Hence the telecom comparison. In the late 1990s Lucent Technologies and Nortel Networks financed their own customers' equipment purchases; revenue looked excellent until borrowers could not repay, after which Lucent's stock fell more than 99% from its peak and Nortel went bankrupt ^[14]. Crypto Briefing is explicit that the cases differ: Nvidia is not lending directly, and the intermediaries are among the most sophisticated institutions in global markets ^[15]. The rhyme is in the mechanism, not the counterparties. Financing that lets buyers take more product than their own balance sheets would support pulls revenue forward and leaves the credit question for later.

Watch three things. Whether the purchase obligation line grows or shrinks in the next quarterly filing. Whether the $4.5 billion of inventory charges proves to be a one-off or the first of a series ^[8]. And who ends up holding the paper if utilisation on financed compute comes in below plan ^[12].