27% a month compounds to 1,660% a year. The guidance says 70%

Jensen Huang repeated Nvidia's guidance at the Goldman Sachs technology conference on Thursday: revenue could grow 70% next year, and he is confident about it, TechCrunch reports. In the same appearance he gave a second growth number, and the two do not live in the same world.

The arithmetic of the guidance first:

• 70% on about $400 billion of expected revenue this year means roughly $680 billion next year.
• The increment alone is $280 billion of new annual revenue, to be found in twelve months.
• 27% a month, compounded across those twelve months, is seventeen and a half times, or about 1,660%.

That second figure is the month-on-month order growth Huang quoted for one product, the system that pairs 36 Grace processors with 72 Blackwell accelerators. It describes a product ramping into its first full year of shipments, not a rate anyone expects to hold. Quoting it beside an annual guidance of 70% invites a reader to average two numbers that measure different things.

The Nvidia revenue growth question is really about the $280 billion, and that is what makes his answer about visibility interesting. He says the company tracks every gigawatt of land, power and building shell on the planet, and that neoclouds, OEMs, clouds and AI-native firms all report back. A supplier with that view is forecasting from an order book rather than from a model.

One GPU, $8.5 million, 250,000 kilowatts

Huang also redefined the product to fit the price. One GPU is no longer $399, he said; it is $8.5 million, two million parts, 250,000 kilowatts, connected with NVLink. Nvidia's own engineering blog said this week that a Vera Rubin rack contains 1.3 million parts, so the two-million figure describes something larger than one rack, and the power figure describes something larger again.

Run the power number on its own. A load of 250,000 kilowatts consumes about 2.19 billion kilowatt hours in a year, which at eight cents industrial is roughly $175 million of electricity. That is twenty times the $8.5 million purchase price, every year. The two figures are not describing the same object, and the sentence invites you to think they are.

Then there is the circularity question, which he answered with a joke.

“Well, it's not circular because we put a little bit of money in, and a lot of money comes back. I look at the spreadsheet, we put in $1 and $100 comes back in.”

— Jensen Huang, Goldman Sachs conference, 10 September 2026

Jensen Huang, Goldman Sachs Communacopia and Technology conference, 10 September 2026

Take the ratio literally against the other number he gave. Huang says he has seen $100 billion of customer contracts behind the companies Nvidia backs. At a hundred dollars back for every dollar in, that implies about $1 billion invested, which is a small fraction of what Nvidia has publicly committed to partners. The hundred to one is a line, not a column in a spreadsheet.

What would slow it down

The honest part of his case sits at the end of the TechCrunch piece. Much of the demand comes from AI-native startups spending raised capital on their own AI use, and efficiency improves as an industry matures. We watched that efficiency arrive this week, when Nvidia's own tuning lifted a server from fourteen concurrent users to forty on the same four accelerators, a straight cut in the cost of inference. Every gain of that kind is hardware someone does not need to buy.

Huang's record on confident numbers is mixed in a specific way. He said AGI has arrived on a stage this month and "for many tasks" on an earnings call, and the difference between those two statements is the difference between a claim and a caveat. The 70% has the same structure: a firm number, a soft verb, and a base year that has not finished yet.

This article is for informational purposes only and does not constitute investment advice.