Jensen Huang repeated Nvidia's forecast of 70 percent revenue growth next year at the Goldman Sachs Communacopia and Technology conference on Thursday, and spent most of his appearance explaining why he is confident rather than hopeful. With analysts putting Nvidia's current fiscal year near $400 billion, that guidance implies roughly $680 billion in revenue — a figure that would require the company to add more in twelve months than it earned in total two years ago.
Key takeaways
- Huang reaffirmed guidance of about 70 percent year-over-year revenue growth, implying roughly $680 billion against an expected $400 billion this fiscal year.
- Orders for the rack combining 36 Grace CPUs with 72 Blackwell GPUs are growing 27 percent month over month, and Huang priced a full connected system at $8.5 million.
- He rejected the circular-deal criticism, saying Nvidia only invests where customer contracts already exist and that he has seen $100 billion worth of such contracts.
The case for visibility
Huang's argument rests less on demand than on information. Nvidia sits between memory suppliers, contract manufacturers, cloud providers, neoclouds and the AI-native startups doing the spending, and he said the company tracks every gigawatt of land, power and data center shell worldwide. Partners report back from each layer, which in his telling means Nvidia knows where capacity is going before anyone else does.
He also pushed back on how the business is understood. The word GPU still evokes a consumer card, and Huang offered new arithmetic: one GPU today is not a $399 part but an $8.5 million system of two million components drawing 250,000 kilowatts, connected by NVLink and shipped by aircraft. The inference and training workloads that matter now consume racks, not cards.
The specific number he volunteered as evidence was 27 percent month-over-month order growth for the Grace-Blackwell rack. Monthly sequential growth at that rate is a far more immediate signal than annual guidance, and it is the kind of figure a CEO offers when he expects to be doubted.
The circularity question he cannot shake
Nvidia invests in companies that then buy Nvidia hardware, an arrangement critics compare to the vendor financing that preceded the telecom bust and the collapse of suppliers like Lucent Technologies. Huang's response at the conference was flippant and pointed at once: it is not circular, he said, because a little money goes out and a lot comes back — a dollar in, a hundred back, and if that is circular, he would like to do more of it.
His substantive defence was about sequencing. Before Nvidia invests, he said, it verifies that the company already holds contracts generating revenue from real customers, and he put the total of such contracts he has reviewed at $100 billion. The distinction he is drawing is between financing demand that exists and manufacturing demand that does not.
Whether that holds depends on where the money underneath originates. As TechCrunch noted in its account of the session, Huang concedes that much of AI's growth now comes from AI-native startups raising enormous sums and spending most of it on their own AI usage. Contracts backed by venture funding are contracts, but they are not the same as contracts backed by operating profit.
What could break the streak
Competition is arriving from every direction at once. Amazon, Microsoft and Google are each building their own accelerators, Anthropic and OpenAI are designing silicon, Cerebras is newly public and startups such as Etched are targeting narrower workloads. None has yet dented Nvidia's position, and Huang's counter is that every lab's models run on his platform regardless of who else builds chips.
The subtler risk is efficiency. Huang acknowledged that AI-native companies are the current growth engine, and those companies are under pressure to spend less per token as they mature. Architectural work aimed at serving larger models on less hardware cuts directly against unit demand, and Nvidia's own ecosystem is producing it. That pressure is already visible in Nvidia's financing posture, including a trimmed data center backstop for OpenAI.
Nvidia has absorbed every such prediction so far. But the guidance Huang reaffirmed on Thursday leaves little room for a soft quarter, and a company forecasting $680 billion is forecasting that nothing in a very complicated supply chain goes wrong.
FAQ
How much revenue would 70% growth give Nvidia?
Analysts expect Nvidia to finish its current fiscal year at roughly $400 billion in revenue. Growth of 70 percent would put next year near $680 billion. Huang first issued that guidance last month alongside another record quarter and repeated it at the Goldman Sachs conference on Thursday.
What are Nvidia's circular deals?
The term describes Nvidia investing in companies that subsequently purchase its hardware, which critics say inflates apparent demand. Huang argues the deals are not circular because Nvidia confirms a company already has revenue-generating customer contracts before investing, and he says he has reviewed $100 billion worth of such contracts.
What is the Grace-Blackwell system Huang cited?
It is a rack-scale computer combining 36 Grace CPUs with 72 Blackwell GPUs, which Huang said is seeing 27 percent month-over-month order growth. He described a fully connected system of this type as costing $8.5 million and containing around two million parts.
