Nvidia now sits on both sides of its own order book

The balance sheet tells this better than the deal list does. Stakes in private companies moved from $22.3bn to $47.9bn in under a year ^[2], an increase of $25.6bn and a multiple of roughly 2.15 ^[15]. Add the $18bn now committed and the position approaches $65.9bn ^[16], about 69 percent of the $96.22bn Nvidia booked in its fiscal second quarter ^[17]. At that size, the portfolio is part of how the results get made rather than a sideline to them.

The reading problem this creates is specific. Nvidia guided fiscal Q3 to about $108bn plus or minus 2 percent ^[11], roughly 12 percent above the prior quarter ^[19], and data center accounted for about 93 percent of Q2 revenue ^[18]. Anyone sizing capacity, land or power off that guide is reading a demand curve in which some of the buyers are also investees ^[1][2]. The trade Nvidia has made is plain enough: it gains claims on the power and site supply that could otherwise stall its shipments ^[3], and it gives up the independence of the demand signal everyone downstream uses to plan.

Nvidia's finance chief took the charge head on. On the earnings call, CFO Colette Kress said that while some describe this as circular financing, "we see it differently," and that measured against the strength of demand, the business created, the ecosystem built on Nvidia's platform and the returns on invested capital, the investments will be excellent and "our risk is limited" ^[12]. That answer addresses pricing, where she may well be right, but it does not address correlation, which is what matters to a developer whose own financing assumes Nvidia-adjacent counterparties stay solvent.

The board-deck version runs like this: the largest customers are building their own chips ^[14], so Nvidia is buying position in the layers around them, and with more than $500bn arranged alongside Wall Street firms ^[4], most of the build is funded by other people. It is incomplete because $500bn is about 28 times the $18bn equity commitment ^[20]. Nvidia's own cash is the mobilising layer rather than the funding layer, which is efficient while utilisation holds and concentrating if it slips. The cheques are real either way: $2.94bn in cash went out on the Groq deal announced in December ^[9], and the arrangement with coding startup Poolside was described as multibillion-dollar ^[5].

What the record does not say is whether any of this reaches commercial terms. Public reporting has not established how, or whether, individual investments change access to Nvidia hardware, cloud capacity, model tooling or pricing for developers ^[7]. So the question for this quarter is narrow: whether an Nvidia-backed counterparty is better credit, or simply the same credit with a louder sponsor. The question for the decade is whether a guide that assumes no China data-center compute revenue ^[11] gets met by demand Nvidia has part-funded. While growth holds, both readings produce identical purchase orders, which is why the distinction has to be drawn before it is tested.