One GPU now costs $8.5 million. That’s not a product line. That’s an infrastructure monopoly with a price tag attached. Jensen Huang made that point Thursday at the Goldman Sachs Communacopia + Technology conference, and it landed exactly the way he intended, as a reminder that Nvidia is no longer the company that sold graphics cards to PC gamers.
According to TechCrunch, Huang used the event to repeat guidance he first gave last month: Nvidia expects revenue to grow 70% next year. Analysts already project the company ends its current fiscal year around $400 billion. So 70% growth puts next year somewhere near $680 billion. That’s not a rounding error. That’s a number that makes most tech companies look like side projects.
The skeptics have a real case. Nvidia faces pressure from nearly every direction. Amazon, Microsoft, and Google are all building their own AI chips in-house. OpenAI and Anthropic are doing the same. Cerebras just went public. Startups like Etched are coming for specific workloads. The question everyone keeps asking is when, not if, some of that demand shifts away from Nvidia silicon.
Huang’s answer is visibility. His argument is that Nvidia is embedded so deeply across the AI supply chain that it effectively sees the whole market before anyone else does. The company tracks data center construction globally, including land, power capacity, and building shells, before a single server gets installed. It gets demand signals from cloud providers, OEMs, AI-native startups, and what Huang calls neoclouds. That’s a level of market intelligence that competitors simply don’t have.
He also pointed to one specific product as a demand signal. A system combining 36 Grace CPUs with 72 Blackwell GPUs is currently growing at 27% month over month. That’s a single product line growing faster than most companies grow in a year.
The circular investment question came up too. Nvidia invests in startups that then buy its hardware, a structure critics compare to the vendor financing schemes that burned companies like Lucent Technologies in the early 2000s. Huang brushed it off with a quip about putting in $1 and getting $100 back. But he also said he reviews real customer contracts before any investment goes out, and that he’s seen $100 billion in such contracts across these companies.
The long-term risk is real. AI infrastructure spending is being driven heavily by AI-native startups burning venture capital. As the industry matures, efficiency improves and token costs drop. That typically means less hardware per unit of output. But right now, demand is outpacing that efficiency curve. And Nvidia, for the moment, is the only company with both the product and the supply chain to meet it at scale.
