Jensen Huang says Nvidia will grow 70% next year. Here is what that number actually means.
The chip giant's CEO told investors he can see the future of AI spending, because almost every company building AI tells Nvidia exactly what they are buying. The math points to roughly $680 billion in revenue by the end of next year.
Key points
- Nvidia CEO Jensen Huang repeated his 70% year-on-year revenue growth forecast at the Goldman Sachs Communicopia + Technology conference on Thursday.
- Analysts expect Nvidia to close its current fiscal year near $400 billion in revenue, putting 70% growth at around $680 billion the following year.
- One Nvidia system combining 36 Grace CPUs with 72 Blackwell GPUs is growing sales at 27% month over month.
- Huang said Nvidia has reviewed $100 billion worth of revenue contracts held by startups it has invested in, and calls those bets low-risk.
- Amazon, Microsoft, Google, Anthropic, OpenAI, and startups including Cerebras are all building competing chips.
Why does Nvidia think it can see the future?
Huang's confidence comes from one simple fact: almost every organisation building AI buys Nvidia hardware and reports its plans back to the company.
He told the Goldman Sachs conference that Nvidia tracks "every single gigawatt of land, power, shell" (the shell being the outer structure of a data centre before computers go in), meaning every such project on the planet. Cloud providers, smaller hosting firms called neoclouds, and AI startups all feed Nvidia information about what they're ordering next. That gives Huang an unusually early look at where spending is headed.
"Nvidia runs every model. Every single lab can use us," he said, naming Anthropic and OpenAI alongside open-weight models, the kind anyone can download and run.
What is Nvidia actually selling at these prices?
Nvidia started life making graphics cards for PC gamers, priced around $399. That framing, Huang argued, no longer fits.
The company's flagship product today is a full computer system: 36 Grace CPUs wired to 72 Blackwell GPUs via Nvidia's own high-speed link technology called NVLink, totalling roughly 2 million parts and drawing 250,000 kilowatts of power. One such system costs $8.5 million. Thousands are shipping right now, with month-over-month sales climbing at 27%.
| Metric | Figure |
|---|---|
| Single system price | $8.5 million |
| GPUs per system | 72 (Blackwell) |
| CPUs per system | 36 (Grace) |
| Month-over-month sales growth | 27% |
| Projected FY revenue | ~$400 billion |
| Projected next-year revenue (at 70%) | ~$680 billion |
Should investors and workers worry about the competition?
Competitors are real. Amazon, Microsoft, Google, Anthropic, and OpenAI are each building their own AI chips to reduce dependence on Nvidia. Cerebras went public specifically to offer an alternative.
Huang's counter-argument is that Nvidia's product isn't just a chip. It's the entire stack: the hardware, the software that runs on it, and the relationships with every major buyer. Switching that out takes years.
One caveat deserves attention, though. Nvidia itself admits a large slice of today's AI spending comes from startups burning investor money on their own experiments. As the industry matures, those companies will almost certainly get more efficient and buy less hardware. Seventy percent annual growth can't continue indefinitely.
Huang's remarks also touched on Nvidia's practice of investing in AI startups that then buy its hardware. Critics have compared it to deals that helped sink telecoms supplier Lucent Technologies in the early 2000s. We reported on 7 September that Nvidia's investment portfolio had already hit $99 billion, with more than $40 billion committed in 2026 alone. Huang brushed the criticism off: "We put in $1 and $100 comes back in," he said, adding that Nvidia only invests after confirming a company already holds real customer contracts worth, in aggregate, $100 billion.
What it comes down to: Nvidia's numbers are genuinely extraordinary right now, and Huang's visibility into the AI supply chain is real. But efficiency gains always catch up with infrastructure booms, and every company named above is working to make sure Nvidia's next $680 billion is harder to collect than the last. The next 12 months look strong. After that, it's a genuinely open race.
