Jensen Huang projects 70% Nvidia revenue growth for next year

Full analysis

Your draft

Jensen Huang stood up at Goldman Sachs and told the room Nvidia will grow revenue 70% next fiscal year, from roughly $400 billion to something near $680 billion. He says he's confident. For anyone building or buying AI, the question isn't whether Nvidia is impressive. It's what a forecast like that tells you about the price and availability of the compute you rent every day, and whether the demand holding it up is real or partly manufactured.

The Skeptic. Watch the circular money. Huang's defense of it is the interesting part: he says he only invests in startups that already show $100 billion in real customer contracts. Read that slowly. A startup with $100 billion in signed contracts doesn't need Nvidia's check. So why is Nvidia writing it? Because the check comes with an implicit deal: buy our hardware. Nvidia funds the buyer who then buys from Nvidia, and the contract gets counted as demand. That's a demand loop with Nvidia on both ends. It works beautifully while every hyperscaler keeps spending. It looks very different the first quarter one of them pauses. The 70% number assumes nobody blinks.

The Safety Lens. Every frontier lab most people trust to do the careful work, Anthropic, OpenAI, Google DeepMind, runs on one company's hardware. That's not a footnote. It means one export-control ruling, one supply shock, or one Nvidia pricing decision moves the entire safety-research frontier at once. The quieter concern is information. Huang says Nvidia can see every major data-center build on earth through its partner network. Nvidia knows who is assembling AI capacity, at what scale, before any regulator does. That's a concentration of knowledge no government has asked to look at. It's been this way for three years, so it feels normal. It isn't.

The Compute Pragmatist. The figure that reframes the whole story: a single GPU cluster now runs $8.5 million, and one product line is growing 27% month over month. Month over month. That tells you Nvidia isn't selling chips anymore. It's selling full racks, cooling, and the NVLink fabric that stitches the GPUs into one machine. The moat moved from chip design, where Cerebras and Google's own TPUs can compete on specific jobs, to systems integration, where nobody else ships at that price and scale yet. The 70% bet is really a bet that no hyperscaler builds enough of its own silicon fast enough to matter in 18 months. Probably right. Not right forever.

The Builder. Nothing here changes your Tuesday except the part that changes everything. GPU allocation stays tight, pricing power stays entirely with Nvidia, and there is no cheaper hardware coming to rescue your inference budget in the next year and a half. So the teams that win are the ones squeezing more throughput out of the GPUs they already have. Waiting for a price cut is a losing strategy. And the circular-investment flywheel means Nvidia is self-selecting who gets served first. If you're not already inside that flow through a major cloud, expect longer queues and worse terms. Plan your capacity like scarcity is permanent, because for your planning horizon it is.

Where they disagree

Two real fault lines here.

The Compute Pragmatist and the Skeptic look at the same $8.5 million cluster and see opposite things. The Pragmatist sees genuine demand: buyers paying full rack prices with no haggling means the return on AI infrastructure still pencils out for them. The Skeptic sees a number Nvidia partly created by funding the buyers. Both can't be fully right. If most of that demand is real hyperscaler capex, 70% holds. If a meaningful slice is Nvidia's own money coming back around, the growth is softer than it looks and the first hyperscaler pause exposes it.

The second split is timing. Everyone agrees the systems moat cracks eventually, when Google's TPUs, Amazon's Trainium, and custom silicon get good enough. The Pragmatist says not within 18 months. The Skeptic says a forecast made from total lock-in is exactly the forecast that gets extrapolated one fiscal year too far.

What it hinges on

The whole call comes down to one thing: is hyperscaler capex still compounding on its own, or is Nvidia's circular investment propping up the top line? The near-term answer is that the money is mostly real. Microsoft, Amazon, Google, and Meta have all guided capex higher, and their spend dwarfs anything Nvidia could seed. The circular piece is a garnish on a very large plate. Appetizer-sized, against a main course of real hyperscaler buildout. For the next year, 70% is arithmetic off buildouts Nvidia can already see in its supply chain. The risk isn't next year. It's the year after, when the same story needs another leg of growth and the easy visibility runs out.

If you're building on this, the thing to de-risk isn't Nvidia's revenue. It's your own dependence on a single vendor with total pricing power. Test whether your inference workloads run acceptably on a TPU or Trainium alternative, even at a penalty. Not because you'll switch tomorrow. Because a credible second option is the only leverage you'll ever have in a procurement conversation with the one company that knows exactly how badly you need the hardware.

Prediction: Nvidia will report year-over-year revenue growth of at least 55% for the fiscal year covered by Huang's guidance, confirmed by the earnings report that closes that fiscal year in early 2027.

Confidence: Medium. The committed buildouts are real, but 70% is a hard number to hit exactly.

Why: Huang isn't guessing; he's reading committed data-center orders through Nvidia's supply-chain visibility, and the four hyperscalers have all publicly guided capex higher, so the demand base for the next fiscal year is largely locked before it's booked. The 27% month-over-month growth on a single rack-scale product line shows the demand isn't price-sensitive yet, which means near-term revenue is a supply-and-shipment question Nvidia largely controls, not a demand question. The opposite outcome, growth falling below 55%, would require a hyperscaler to actually cut orders inside 18 months, and none has signaled that. I'm landing below Huang's own 70% because that's the number a CEO gives at a bank conference, and the real range of outcomes sits under the confident headline.

Revisit by 2027-04-30: We're right if Nvidia's reported full-year growth for the guided fiscal year lands at 55% or higher. We're wrong if it comes in below 55%, whether from a capex pullback, an export-control shock, or hyperscaler silicon eating share faster than expected.