Nvidia has spent three years training Wall Street to expect the impossible and then beating it anyway. On Wednesday it did something genuinely new: it told investors what to expect a full year out.

The numbers everyone came for were, as usual, absurd. Revenue for the quarter ended July 26 hit $96.2 billion, up 106% year over year and roughly $4 billion past the $92.2 billion analysts had modeled. Net income reached $59.7 billion, up 126%. Adjusted earnings landed at $2.22 a share against expectations of $2.06 to $2.09. Data center revenue came in at $89 billion, up 117%.

None of that was the news. Buried in the same release was a preliminary revenue outlook for fiscal 2028: growth of approximately 70%. Nvidia has never issued a year-ahead forecast as a public company. Analysts had penciled in about $570 billion for fiscal 2028, or 44% growth. Seventy percent implies closer to $690 billion to $700 billion, more than $100 billion of revenue consensus had not accounted for. Shares dipped on the print, then reversed and climbed more than 4% after hours.

Breaking precedent on purpose

"We wanted to make sure that everybody has the same set of information," founder and CEO Jensen Huang said on the call. "We have got a huge year coming up next year, and it is going to be pretty extraordinary." He named the break with practice himself: "It is the case that we have never forecasted, never guided to a year in advance."

Melissa Otto, global head of Visible Alpha research at S&P Global, said the magnitude blew away expectations. "I think what wowed the market was that 70% fiscal year 2028 number that was way ahead of Visible Alpha consensus," she said. "The whole market was like, whoa, 70%."

The critical detail: 70% is not a demand figure, it is a supply figure. CFO Colette Kress said customer forecasts point to demand roughly doubling next year; Nvidia is guiding to what it can physically build. "Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%," Huang said.

For the current quarter, Nvidia guided to roughly $108 billion, plus or minus 2%, with zero China data center compute revenue included. The mix matters: hyperscalers accounted for $48.7 billion, up 102%; AI clouds, industrial and enterprise customers hit $40.3 billion, up 138%. That non-hyperscale crowd, Huang said, "represents about half our business, and that is growing 100% a year."

The balance sheet is now part of the product

Nvidia has become an active financier of its own customers, and Kress preempted the obvious objection before an analyst could raise it. "We recognize the scale of this support, and we know some will call this circular financing," she said. "We see it differently."

The disclosed scale is not small. Nvidia reported maximum gross guarantee exposure of $108.5 billion, most of it credit support for SB Energy's Ohio campus, which will host Nvidia compute leased to OpenAI, plus $3.5 billion backing AI cloud partner leases. It has separately invested nearly $50 billion in frontier AI labs and partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than $500 billion in third-party capital.

Kress argued the structure is misunderstood. "Independent capital still underwrites every deal on its own merits. We are not making loans," she said, describing an arrangement in which Nvidia guarantees a minimum share of a neocloud's capacity to satisfy lenders, then takes a cut of rental revenue above that threshold. "In this model, we get paid twice."

That is a real answer, and an incomplete one. Kress said frontier labs will account for roughly a quarter of Nvidia's business next year. Nvidia is underwriting demand it then books as revenue, and the credit market has noticed. Bill Birmingham, managing director at Rex Financial, noted that when July reports surfaced that Nvidia might guarantee up to $250 billion of OpenAI capacity in Ohio, credit default swaps repriced its five-year risk from 40 basis points to 82. "The equity shed $250 billion in turn," Birmingham wrote in a pre-earnings note. "Even though the final number came in at $105B, the market read this as less demand and not less risk."

The memory bill comes due

The second pressure point is simpler. Nvidia's long-term supply commitments exploded from $119 billion to $279 billion in a single quarter, driven, in Kress's words, primarily by memory procurement. Up to $160 billion of that is memory alone, including a supply pact with SK hynix.

That shows up in margins. Gross margin was 75.0% in the quarter. Nvidia guided to about 74% for Q3, said margins will bottom at 71% to 72% in Q4, then recover to 72% to 73% next fiscal year as price increases take effect. The company has reportedly warned large customers of roughly 15% price hikes on AI servers.

Huang framed this as foresight. "A long time ago, people asked me why it is that we are working with memory suppliers when we are a chip company," he said. "Today, people understand it is really quite genius that we were working on our supply chain so far upstream."

Birmingham was less charmed. "It is dangerous to raise prices when ROI for AI at the customer level is still unknown," he wrote. Otto's counterpoint: the market had modeled 72.6% for Q3, so 74% reads as resilience.

What to watch

Three markers. First, whether 70% proves a floor or a ceiling. Nvidia has made supply the explicit constraint, so HBM contracts and packaging capacity are the leading indicator now, not order books. Second, whether the $108.5 billion guarantee figure grows and whether credit spreads widen with it. The equity market rewarded this quarter; the credit market has been skeptical for months. Third, the Q4 margin trough. If it lands below 71%, the pass-through thesis has failed and Nvidia is eating memory inflation rather than exporting it.

And one wildcard outside the model entirely: China contributes nothing to guidance. Whatever happens there is upside Nvidia has declined to promise.