CEO Jensen Huang keynote Blackwell Vera Rubin
On: Nvidia’s At the annual developer conference on Monday, CEO Jensen Huang took the stage and said he expects purchase orders between Blackwell and Vera Rubin to reach $1 trillion by 2027.
Last year the company had estimates of a $500 billion revenue opportunity between the two chip technologies. Following Nvidia’s earnings report last month, Finance chief Colette Kress said the company expects growth this year to exceed that forecast.
Huang said demand is increasing from both startups and large companies. Nvidia shares rose nearly 2% on Monday.
“If they could get more capacity, they could produce more tokens and their revenue would increase,” Huang said at GTC in San Jose, California.
Nvidia’s graphics processing units for AI have turned the brand into a household name and the world’s most valuable publicly traded company, valued at approximately $4.5 trillion. As mass adoption of AI progresses from chatbots to intermediary applications that leverage other agents to perform tasks, the number of tokens created has exploded, further increasing the need for inferences to be executed at higher speeds.
The chipmaker said in February that annual revenue this quarter would rise about 77% to about $78 billion. The company has reported revenue growth of over 55% for 11 consecutive quarters.
Nvidia is scheduled to release Vera Rubin later this year. The company claims that the system, consisting of 1.3 million components, will deliver 10 times more performance per watt than its predecessor, the Grace Blackwell. This is an important development considering that energy consumption is one of the most critical problems facing artificial intelligence development.
Also on Monday, Huang unveiled the Nvidia Groq 3 Language Processing Unit, or LPU, the company’s first chip at the startup, which it acquired mostly through a $20 billion asset purchase in December, its biggest deal to date. It is expected to be released in the third quarter.
Founded by the creators of Groq Google’s An in-house tensor processing unit that has gained traction in recent years as a rival to Nvidia’s graphics processing units. Groq 3 LPU is built to enhance the GPU’s technology with an optimized core to accelerate it.
Huang unveiled an entire rack dedicated to housing the new Groq accelerators.
The Groq 3 LPX rack will carry 256 LPUs and will sit alongside the Vera Rubin rack-scale system that will ship to customers later this year. Huang said the Groq LPX rack can increase the token-per-watt performance of Rubin GPUs by 35x.
“We put together two very different processors, one for high throughput and the other for low latency. That still doesn’t change the fact that we need a lot of memory,” Huang said. “And so we’re going to add a bunch of Groq chips that increase the amount of memory it has.”
Huang also showed off a prototype of Kyber, Nvidia’s big rack architecture breakthrough after Rubin. It will integrate 144 GPUs into compute trays placed vertically rather than horizontally to increase density and reduce latency. The Kyber design will be available in Nvidia’s next rack-scale system, the Vera Rubin Ultra, expected to be released in 2027.
About two hours into his keynote, Huang turned to the OpenClaw phenomenon started in January by Austrian software developer Peter Steinberger. It has grown in popularity in part due to attention on social media, as consumers and businesses flock to products that can autonomously complete tasks, make decisions, and perform actions on users’ behalf, without constant human guidance.
Steinberger joined OpenAI last month, and CEO Sam Altman said OpenClaw “will live on a foundation as an open source project that OpenAI will continue to support.”
Huang highlighted a new developer toolset that will help people build and experiment with what’s possible in new areas of AI using Nvidia hardware. Specifically, he introduced a reference stack for OpenClaw called NemoClaw and helped make it “enterprise ready”.
“It finds OpenClaw and downloads it. It creates an AI agent for you,” Huang said.
— CNBC’s Jordan Novet contributed to this report.
WRISTWATCH: Inside Nvidia’s Vera Rubin AI system
