Loading today’s stories
Leadership2 distinct publishers3 min readPublished
Private holdings ran from $22.3bn to $47.9bn in under a year, with another $18bn committed. Anyone reading Nvidia's demand signal is partly reading capital Nvidia itself supplied, though its CFO rejects that framing.
The Board Room · Leadership desk
Compiled by The Board RoomSomething wrong?How this is made
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 [1]. Add the $18bn now committed and the position approaches $65.9bn [2], about 69 percent of the $96.22bn Nvidia booked in its fiscal second quarter [3]. 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 [8], roughly 12 percent above the prior quarter [5], and data center accounted for about 93 percent of Q2 revenue [4]. 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" [10]. 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 [7]. 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 [6], 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 [13]. 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 [8] 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.
Ranked by verification strength, evidence, and original report placement.
Nvidia said in its earnings report that it has $18 billion committed to equity investments for the rest of its fiscal year, focused on AI model makers, infrastructure financiers, and other private companies.
Nvidia held $47.9 billion in private companies as of late July, more than double the $22.3 billion it held at the end of the previous fiscal year.
In August Nvidia announced it had invested $1.5 billion in SB Energy and taken a stake in Cloverleaf Infrastructure, two companies helping to solve the land and power crunch posed by data centers.
Nvidia teamed up with Wall Street firms to raise more than $500 billion in outside funding for AI infrastructure.
Nvidia struck a multibillion-dollar deal with coding startup Poolside and is reportedly eyeing investments or deals with AI search startup Perplexity and Korean inference chip startup Rebellions.
Lango said licensing technology, hiring talent, and taking minority stakes can help avoid some of the regulatory scrutiny and integration headaches that can accompany an acquisition.
Distinct publishers with included, body-backed reporting in this cluster.
businessinsider.com
1 article · August 26, 2026
letsdatascience.com
1 article · August 29, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Nvidia booked $40B of its $89B data center quarter outside the hyperscalers1 distinct publisher
invest
Nvidia's August 26 print: 92% of the quarter rides on one segment1 distinct publisher
invest
Nvidia's pre-earnings re-rating is a credit story, not a chip story1 distinct publisher
product
Nvidia circles Rebellions because the low-power inference tier is not optional2 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One issuer, one reporting channel
Every dollar figure here — $18 billion forward, $47.9 billion held, $2.94 billion paid on Groq — traces to Nvidia's own earnings report and call, relayed by Business Insider. Let's Data Science checks none of it; it credits the deal facts to Business Insider and the revenue lines to CryptoBriefing. Company disclosure of this specificity is usually dependable, which is why the score sits well above the middle, and nobody has independently verified a number in it, which is why it stops there. The deal-by-deal detail is softer still: Poolside is "multibillion" with no figure, Perplexity and Rebellions are only "reportedly" in play.
Cash has already moved
This is not an announcement waiting on execution. $1.5 billion is in SB Energy, $2.94 billion of cash has left the building for Groq, and the private-holdings line more than doubled inside a year. What has not shown up is any consequence you could observe from outside: Let's Data Science says plainly that no reporting establishes whether the stakes shift hardware access, cloud capacity, tooling or commercial terms. Deployment of capital, heavy; downstream effect, unmeasured.
Framing runs slightly ahead of the arithmetic
The reporting is more careful than the story it implies. Two numbers carry the circularity suggestion: the pro forma $65.9 billion position, about 69% of one quarter's revenue, and the $500 billion of outside funding Nvidia helped arrange, which is other people's money at roughly 28 times its own commitment. Neither tells you what share of Nvidia's revenue comes from companies Nvidia funded, and that is the question the charge actually rests on. Business Insider at least gives Kress room to reject the framing; Let's Data Science leaves her out, which widens the gap on its side.
Every interpretation is supplied by an interested party
Three voices, three positions. Nvidia's CFO is answering a circular-financing charge on an earnings call, where "our risk is limited" is the only available answer. The countervailing quote comes from a chief technology analyst at InvestorPlace, whose business is selling exactly this kind of call on exactly this stock. And Nvidia decides which commitments get a number and which stay "multibillion." None of that makes the figures wrong; it does mean both readings of them come from people with something riding on the outcome.
Firm on the figures, thin on what they buy
The balance-sheet and revenue numbers are specific, dated and unlikely to move. What the money purchases — control, priority supply, pricing — goes unaddressed by both publishers, and since the second is largely a restatement of the first, a single error would propagate uncorrected. Deliberately held below where the arithmetic alone would put it.