Nvidia Confirmed the AI Boom. September Decides Who Can Still Finance It.

Wednesday's print was the best report card the AI trade has ever received, and the stock market's reaction lasted about a day.

Nvidia reported fiscal-second-quarter revenue of $96.2 billion, up 106% from a year ago, with data-center sales of $89 billion, up 117%. Management guided the current quarter to roughly $108 billion — another ~12% step up — and non-GAAP earnings of $2.22 per share came in ahead of the roughly $2.10 analysts expected against $92.2 billion of consensus revenue. Shares jumped more than 4% after the close, powered higher again the next session, then faded about 4% by Friday afternoon with Fed Chair Kevin Warsh speaking at Jackson Hole. A record quarter, and half of the post-earnings pop already given back.

The quarter did its job: it retired the demand-deceleration question. There is no deceleration. Management had already pointed to roughly $1 trillion in combined Blackwell and Vera Rubin sales through 2027, and the Vera Rubin ramp is now the growth engine. The interesting number is therefore not the beat. It is the fact that an 85-point quarter against an 80-point hurdle no longer buys a durable move in the stock.

The deeper question is why. Nvidia enters this week at a $5.3 trillion market cap and roughly 17 times trailing sales; Broadcom trades near 24 times. These are long-duration assets, meaning their value sits in earnings years away — which is exactly what a rising price of money hits hardest. That is the lens for the reaction pattern of the last two months, and it explains why three companies delivered what the bull case demands and got sold for it:

• Broadcom reported $10.8 billion of AI chip revenue in a single quarter, up 143%, beating every estimate — and the stock fell 12.6% the next day when management refused to raise its AI outlook again.
• Oracle reported a record fiscal 2026 with $638 billion of remaining performance obligations, the signed-contract backlog for future cloud revenue — and the stock fell about 10% in the sessions that followed.
• Alphabet raised its 2026 capex forecast to a $205 billion top end — and lost 7%, because free cash flow turned negative in the quarter and long-term debt more than doubled over six months to $98 billion.

Three independent confirmations of real, contracted AI demand, and three negative reactions. That is not noise; it is a hurdle that moved. The old question was whether the industry could build AI infrastructure fast enough. The new question is at what cost of money, and who carries the balance sheet while it does.

Look at where the constraint has migrated. In 2023 and 2024 the bottleneck was accelerators. Nvidia's own release now runs the list well past chips: a deal with SK hynix to lock down next-generation memory, a land-and-power partnership in Ohio, and financing. Two weeks before the quarter, Nvidia lined up partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion in third-party capital for AI data centers, and it has been reported to offer credit support to AI cloud providers in exchange for a share of revenue. The scale is the point: the five largest U.S. cloud and AI infrastructure spenders have guided to roughly $660 billion to $690 billion of capex for 2026, nearly double 2025 levels — money that now has to be financed at whatever interest rate the Fed picks. In other words, the buyer's cost of capital has started showing up inside Nvidia's own income-statement risk.

The Federal Reserve has noticed. Its July minutes cite "AI-related demand" — alongside Middle East supply shocks and tariffs — among the forces driving price increases. On July 29 the committee held rates at 3.50% to 3.75% in a 9-3 vote, with three members dissenting in favor of a hike. Since then the 30-year Treasury has reached its highest yield since 2007, and the odds of a September hike have swung between roughly a third and a half over the past month — with nearly four in five futures traders now expecting at least one hike by year-end. For the negative-free-cash-flow names that carry most of this buildout, a hike is not a headline; it is a direct increase in the price of the product they are buying.

September is where the two stories finally meet, in four checkpoints:

1. September 2 — Broadcom. It guided the quarter to roughly $29.4 billion of revenue and $16 billion of AI revenue, more than 200% higher than a year earlier. The test is whether a print that hits or beats that guidance holds the stock after June's sell-the-news verdict. If it does, the hurdle is easing. If it doesn't, valuation — not demand — is what the market is repricing.
2. September 11 — August CPI. July's 3.4% year-over-year read came in close to expectations and took some urgency out of a hike. This print, a week before the decision, is the last swing factor the committee sees, along with the September 4 jobs report.
3. September 14 — Oracle. The test here is conversion: whether the $638 billion backlog turns into revenue and margin fast enough to keep the balance sheet from becoming the story.
4. September 15-16 — the FOMC. A hike — or a hawkish signal — raises the discount rate on every long-duration AI multiple at once. A hold with no added threat lets the earnings clock dominate again.

The so-what: the AI earnings story is not broken; it is more confirmed than ever. What has changed is what drives the next price move. The rerating trigger is a September Fed that hikes — or threatens to — while no hyperscaler trims a multi-year capex plan and Broadcom's backlog keeps converting. That would tell you the fundamental flywheel of contracted demand outweighs the discount rate, and that the recent pullback is valuation washing out rather than the cycle ending. The signal that breaks the narrative is the first hyperscaler that answers higher rates by cutting capex, or a financing window that narrows on those $500 billion platforms. One trim is a signal. A second is a regime change.

We only need to watch A: whether any major capex commitment gets trimmed in response to the price of money. The margin and revenue consequences follow from it. Between now and the Fed's decision, that is the metric that decides which September story — the earnings clock or the discount rate — is actually running the market.