Seventy percent. Confidently.
That’s the number Jensen Huang put on Nvidia’s next fiscal year, speaking Thursday at Goldman Sachs’ Communacopia + Technology Conference. His exact words: “I think we could grow 70% year over year. We’re confident about that.” He’d said it before. He said it again.
Analysts have Nvidia finishing its current fiscal year around $400 billion in revenue. Apply 70% and you land near $680 billion. That is the projection. Not a stretch goal, not a bull case buried in a footnote. The CEO’s stated expectation.
I review AI tools for a living. I spend my days telling people that the agent they’re excited about is a wrapper around an API call with a nice loading spinner. So my instinct when a CEO says “70%” is to reach for the skepticism. But the skepticism has to be pointed at the right thing.
What Huang actually said versus what people heard
His reasoning came down to two things: Nvidia’s position in the AI market and demand for its products exceeding what it can supply. Demand for AI computing continues to outrun availability. That’s the whole argument. It’s not complicated, and it’s not a technology claim. It’s an order-book claim.
That distinction matters more than most coverage lets on. Huang isn’t predicting that AI will get smarter next year. He’s predicting that the people buying his chips will keep buying his chips. Those are separate bets that happen to be sitting in the same sentence.
Supply constraint came up too. When you’re supply-constrained, forecasting gets easier in one direction and harder in another. Easier because you know demand exceeds what you can ship, so the ceiling is your own manufacturing. Harder because your growth is now a function of how much you can physically produce, and that involves partners, fabs, packaging, and power.
Why this matters to anyone building with AI
Here’s my angle, and it’s the one I care about as someone who tests these tools. Nvidia’s growth forecast is the clearest signal we have about what happens to the cost of running AI over the next year or two.
If Nvidia grows 70% because it’s shipping more units at a similar price, compute gets more available and inference costs drift down. That’s good for anyone building an agent that needs to make forty model calls to answer one question. If it grows 70% because prices hold high in a supply-starved market, the economics of AI products stay ugly for everyone who isn’t a hyperscaler.
Huang didn’t break that down, and I’m not going to pretend to know which it is. But it’s the question worth asking when you see the headline number, rather than treating $680 billion as a scoreboard.
What I’d want to know
- How much of that growth is locked into existing commitments versus expected new demand
- Whether supply constraints ease enough to change per-unit pricing
- How much of the demand comes from a handful of buyers versus a broad base
- What happens to the forecast if any large customer slows spending
None of that is in the public reasoning Huang gave. He gave dominance and demand. Those are real, but they’re also the two things a CEO says when the specifics are commercially sensitive.
The bubble question
Talk of an AI bubble came up around this same set of remarks, alongside Nvidia beating its Q2 expectations. I’ll say what I actually think: the bubble debate and Nvidia’s revenue forecast are not the same conversation, even though people keep merging them.
Nvidia can hit $680 billion in a year where half the AI startups buying compute never find a business model. Revenue recognized is revenue recognized. Selling shovels works whether or not there’s gold in the hill. The interesting failure mode isn’t Nvidia missing its number next year. It’s what the year after looks like if the buyers stop being able to justify the spend.
Huang is forecasting one year out. That’s a reasonable horizon for someone with visibility into orders. It’s not a statement about whether the whole thing holds together in 2029, and he didn’t claim it was.
My read
I take the 70% seriously because Huang has repeated it, publicly, more than once, on the record, at an investor conference. That’s not a throwaway. Companies don’t casually restate growth targets of that size.
What I don’t take seriously is the framing that this settles anything about AI’s value. A chip company selling out its production capacity tells you demand exists right now. It tells you nothing about whether the products built on those chips are any good. That part is still my job, and the tools I test are still mostly not good.
Two things can be true. Nvidia grows 70%. Your AI agent still can’t reliably book a flight.
đź•’ Published: