Jensen Huang: Nvidia Revenue Could Hit $680B on 70% Growth in 2027

Nvidia CEO Jensen Huang told the Goldman Sachs Communacopia + Technology conference on September 10, 2026, that he expects revenue to grow 70% year over year into next fiscal year. With analysts projecting the current year to close near $400 billion, that puts the next target at roughly $680 billion. Huang credited Nvidia's visibility into every layer of AI infrastructure, from memory chip suppliers to data center shell construction globally, as the basis for his confidence, repeating guidance the company first issued the prior month after another record quarter.

What happened

Metric	Figure
Revenue growth forecast (next fiscal year)	70% year over year
Estimated current fiscal year revenue	~$400 billion
Implied next-year revenue	~$680 billion
Cost of one GPU system (36 Grace CPUs + 72 Blackwell GPUs)	$8.5 million
Parts in that system	2 million
Power draw	250,000 kilowatts
Month-over-month sales growth for that product	27%
Investment-backed contracts Huang says he has reviewed	$100 billion

Huang used the Goldman Sachs conference to push back on the common picture of Nvidia as a chip maker. “Most people think Nvidia builds a chip. I mean, you need airplanes to ship what we build,” he said. A single GPU system today costs $8.5 million, incorporates 2 million parts, and draws 250,000 kilowatts of power. Nvidia ships thousands of these units.

The product he highlighted specifically combines 36 Grace CPUs with 72 Blackwell GPUs, connected through NVLink. That system is growing at 27% per month in sales, according to Huang.

Why does Nvidia expect 70% growth next year?

Huang’s argument rests on two things: ubiquity and visibility. He says every major AI model, including those from Anthropic, OpenAI, and Google, as well as open-weight models, runs on Nvidia hardware. “Nvidia runs every model. Every single lab can use us,” he told the conference.

Beyond the labs, Nvidia works with cloud providers, OEMs, and what Huang calls “neoclouds,” a term for the newer generation of AI-focused cloud operators. Through those relationships, he claims Nvidia is tracking “every single gigawatt of land, power, shell around the world.” A data center shell is the physical building before it is fitted with servers and networking. That level of pipeline visibility is the core of his conviction on the 70% forecast.

He also addressed questions about Nvidia’s practice of investing in companies that then purchase its hardware, sometimes called circular deals. Huang was blunt: “We put a little bit of money in, and a lot of money comes back.” He said Nvidia only invests after confirming a company has real revenue-generating contracts with customers, and that he has personally reviewed $100 billion worth of such contracts. “I’m not taking any risks. I need a sure thing,” he said.

Why it matters

A move from $400 billion to $680 billion in a single year would be one of the largest absolute revenue increases any company has ever posted. For context, the entire global semiconductor industry generated roughly $600 billion in revenue in 2024, so Nvidia alone would be approaching that total from just AI-related hardware.

Competition is real. Hyperscalers including Amazon, Microsoft, and Google are all building custom AI chips. Anthropic and OpenAI are reportedly doing the same. Cerebras recently went public, and startups like Etched are targeting specific AI workloads. Huang’s answer to all of them is that Nvidia is not just a chip: it is a platform, a supply chain, and a visibility layer that competitors cannot replicate quickly.

Huang also acknowledged that much of today’s AI spending is coming from AI-native startups burning investor cash on their own AI operations. As the industry matures, infrastructure spending patterns will likely shift toward efficiency rather than raw scale. That is the single biggest risk to the forecast that Huang did not fully address.

For businesses evaluating AI infrastructure costs, the numbers Huang cited are a useful anchor. If a single GPU system starts at $8.5 million, meaningful on-premise AI capability is still firmly in large-enterprise territory. Most small and mid-size businesses will continue to access AI through APIs and cloud services rather than owned hardware. Understanding how to get value from those tools is exactly where AI integration work at the application layer pays off.

Our take

Huang is a skilled narrator, and the story he tells is internally consistent. Nvidia genuinely does touch every layer of the AI stack, and that visibility is a real competitive advantage, not just a talking point. The 27% month-over-month growth figure on a specific product line is the most concrete signal in the speech, and it is hard to dismiss.

That said, the circular investment structure deserves more scrutiny than his quip gave it. The argument that “we checked and they have contracts” does not fully resolve the conflict of interest. Lucent Technologies ran a similar playbook in the telecom buildout of the late 1990s and it worked until it didn’t.

The 70% growth call is aggressive. For it to land, the AI spending wave needs to hold through the end of 2027. We cover the competitive pressure from AI labs building their own chips regularly in our AI news coverage. Watch whether hyperscaler custom silicon starts showing up in their earnings as a meaningful substitute for Nvidia purchases. That is the metric that will validate or undermine Huang’s thesis before the fiscal year closes.

What to do about it

1. Note the $8.5 million per-system figure as your benchmark when evaluating any vendor claim about “AI hardware costs.” Anything far below that is a different class of hardware.
2. Track hyperscaler earnings calls over the next two quarters for mentions of custom chip adoption rates versus Nvidia GPU purchases.
3. If you are spending on AI cloud services, audit which underlying hardware your provider is using. Pricing shifts at the infrastructure level eventually flow through to API costs.
4. If you are a business owner trying to get ROI from AI tools now, focus on workflow automation and integration at the software layer rather than hardware. Workflow automation built on existing models delivers value without requiring you to bet on which chip wins.

Source: TechCrunch · AI