Nvidia Heads Into Earnings Financing Its Own Demand as $1.8T in Hidden Leverage Builds
"This is not circular financing," Jensen Huang said, and then estimated OpenAI could generate $600 billion of Nvidia revenue through 2030.
That figure is 5.7 times the maximum backstop Nvidia just put on the line. It is also the entire argument. Strip it away and the denial has nothing underneath it but a forecast the company has not yet earned.
Strip away the $600 billion forecast and the denial has nothing underneath it but a projection the company has not yet earned.
Here is the tell. On August 17, Nvidia unveiled a commitment larger than its entire cash pile, and the stock closed at $225.01, down 0.07%, on 93.6 million shares — 64% of its 65-day average. A market so certain of the August 26 beat it couldn't be bothered to price the guarantee.
Last week, we tracked Huang's $500 billion chip-backed securities pact — GPUs reframed as an asset class, financed by Wall Street, with Nvidia backstopping a quarter of every loan. This week the abstraction gets specific: Nvidia has moved from securitizing chips in general to directly guaranteeing one customer's demand in Ohio.
Let me concede the bull case in full, because it is strong.
The beat is nearly a foregone conclusion. Nvidia guided to $91 billion; Bank of America's Vivek Arya models $94–95 billion, and expects a Q3 guide of $107–108 billion against a $104 billion consensus. That is not one surprise. That is the setup for a multi-quarter upgrade cycle.
The revenue underneath the financing is real. Anthropic reportedly told investors its annualized run rate hit $65 billion in July — seven times a year earlier. OpenAI's sits near $40 billion. This is not fabricated demand.
And Nvidia can plainly afford the cash outlays. Free cash flow rose 18-fold in three years to $48.5 billion a quarter. Q1 revenue grew 85% to $81.6 billion, Data Center 92% of it, at a 74.9% GAAP gross margin. The company raised its dividend from a penny to 25 cents, added an $80 billion buyback, and still pledged to return half its free cash flow.
So the objection is not that Nvidia is running out of money. It isn't. The objection is what sits off to the side of the income statement, uncounted.
Now the turn. Count the contingent exposure.
Start with Ohio. Nvidia committed up to $105 billion to backstop lease, power, and a "specified residual-value commitment" on OpenAI's data-center campus at PORTS-Pike, opening between 2028 and 2030 under a 20-year lease.
That ceiling equals 2.09 quarters of Nvidia's latest operating cash flow of $50.3 billion. Sit with that: more than two full quarters of everything the company generates, pledged against a single customer's obligations.
Worse, it exceeds Nvidia's liquidity. Cash and marketable debt securities total $50.3 billion. Add $30.2 billion of marketable equity and you get $80.6 billion. The backstop is 1.30 times that combined pool — and 2.09 times the core cash-and-debt figure alone.
Read plainly: a full guarantee call could exceed Nvidia's current core liquidity. The company could cover it over time from cash flow, but not from what's on the balance sheet today.
Then add the $500 billion pact. Nvidia holds the option to backstop 25% of every GPU loan — roughly $125 billion — treating chips, in Huang's framing, as an asset class "the way banks treat real estate."
Then add the equity. Nvidia holds $30.2 billion in marketable equity securities, up from $12.9 billion a year earlier, including a $30 billion position taken in OpenAI in February. That is $30.2 billion in stakes in the very customers buying its chips.
Stack it: $105 billion in Ohio backstops, up to $125 billion in loan guarantees, $30.2 billion in customer equity. The exposure is concentrated in a small set of frontier labs whose credit profiles Huang himself describes as inadequate.
His own words: many labs "are growing faster than their balance sheets and long-term credit profiles can support" and "lack the decades-long infrastructure contracts and investment-grade financing capacity" to secure capacity alone.
That is the whole rationale for the backstop — and it is also, precisely, the risk. Nvidia is guaranteeing the obligations of customers it has already told you can't guarantee their own.
Which returns us to Huang's denial. "This is not circular financing" holds only if OpenAI's business services its commitments without Nvidia's continued support. The $600 billion revenue estimate is the load-bearing beam — but it is contingent on full deployment and repeated upgrade cycles through 2030.
If OpenAI's build commitments outrun its revenue, the exposure does not vanish. It lands back on Nvidia's balance sheet — the very asset the strategy exists to protect.
Wall Street is choosing to price this at zero. Cantor called the structure "elongated and durable" rather than circular. S&P's 61-analyst consensus is Strong Buy at $302.83, 34.6% above the current price. The published target range runs $180 to $500 — a spread that tells you nobody actually agrees on the risk.
Michael Burry, per Forbes, called parts of the structure a "Wall Street stunt." Paul Meeks of Freedom Capital Markets notes competition is already eating Nvidia's "outrageous margins." Google booked TPU system revenue this quarter; AMD's data-center revenue more than doubled. The technology moat is narrowing, which is exactly why the capital moat is being built.
So here is the specific test for August 26, and it is not the headline number.
The revenue will beat. The Q3 guide will likely be raised. None of that resolves the question that actually matters, which lives in the footnotes.
Watch whether the 10-Q finally quantifies the guarantee exposure — the Ohio residual-value commitment, the 25% loan backstop, the take-or-pay terms — as measurable contingent liabilities rather than narrative.
Until it does, the market is valuing at zero a set of obligations that, if called together, exceed Nvidia's entire current liquidity pool. That is not a bet on whether Nvidia beats. It is a bet on whether the demand Huang is forecasting through 2030 actually shows up to retire the guarantees he has already written.
The beat is priced in. The backstops are not. Read the footnotes before you read the headline.