Nvidia’s CEO Jensen Huang delivers a keynote speech at Nvidia’s GTC conference on March 16, 2026 in San Jose, California. Nvidia’s GTC conference will focus on the latest developments and future uses of AI.
Benjamin Fanjoy | Getty Images
At Nvidia’s At Monday’s annual developer conference, CEO Jensen Huang took the stage to a packed house and said that purchase orders between Blackwell and Vera Rubin are expected to reach $1 trillion by 2027.
Last year, the company had projections for a $500 billion revenue opportunity between the two chip technologies. Following Nvidia’s earnings report last month, finance chief Collette Kress said the company expects growth this year to exceed what was included in that estimate.
Demand is increasing from startups and large companies, Huang said. Nvidia shares rose nearly 2% on Monday.
“If they can just get more capacity, they can generate more tokens, their revenue will increase,” Huang said at GTC in San Jose, California.
Nvidia’s graphics processing units for artificial intelligence have turned the brand into a household name and the world’s most valuable public company, worth about $4.5 trillion. As mass AI adoption has shifted from chatbots that spawn other agents to accomplish tasks to agentic applications, the number of tokens generated has exploded, creating a greater need to run inference at a faster pace.
The chipmaker said in February that revenue would rise about 77% year over year to about $78 billion in the quarter. The company has reported 11 straight quarters of revenue growth of more than 55%.
Nvidia is slated to roll out Vera Rubin later this year. Comprising 1.3 million units, the company claims the system delivers 10 times more performance per watt than its predecessor, Grace Blackwell. That’s a significant development when energy consumption is one of the most critical issues facing AI build-out.
On Monday, Huang unveiled the Nvidia Groq 3 language processing unit, or LPU, the company’s first chip from a startup largely acquired through a $20 billion asset purchase in December, its largest deal to date. It is expected to ship in the third quarter.
Nvidia CEO Jensen Hwang delivers the keynote address at the company’s annual GTC developers conference on March 16, 2026 at the SAP Center in San Jose, California.
Josh Edelson | AFP | Getty Images
Grok is founded by creators of Google An in-house tensor processing unit, which has gained traction in recent years as a competitor to Nvidia’s graphics processing units. The Groq 3 LPU is built to maximize its technology, with one core optimized to accelerate the GPU.
Huang introduced an entire rack dedicated to housing the new Grok accelerators.
The Groq 3 LPX rack can hold 256 LPUs and is intended to sit next to a Vera Rubin rack-scale system that will ship to customers later this year. Huang said the GROC LPX rack can increase the tokens per watt performance of its Rubin GPUs by 35 times.
“We combined, integrated two processors of extreme differences, one for high throughput, one for low latency. It still doesn’t change the fact that we need a lot of memory,” Huang said. “So we’re going to add a whole bunch of GROC chips that expand the amount of memory it has.”
Huang showed a prototype of Kyber, Nvidia’s next big rack architecture leap since Rubin. It integrates 144 GPUs into compute trays that sit vertically instead of horizontally to increase density and lower latency. The Kyber design will be available in the Vera Rubin Ultra, Nvidia’s next rack-scale system, expected to ship in 2027.
In his roughly two-hour keynote, Huang turned to the OpenClaw phenomenon, which was started in January by Austrian software developer Peter Steinberger. It has grown in popularity in part because of the focus on social media as consumers and businesses move toward products that spontaneously complete tasks, make decisions, and take actions on behalf of users without constant human guidance.
Steinberger joined OpenAI last month, and CEO Sam Altman said OpenClaw “will live on in the foundation as an open source project that continues to support OpenAI.”
Huang highlighted a new developer toolkit that lets people use Nvidia hardware to build and experiment with what’s possible in new areas of AI. He introduced a reference stack called NemoClaw, specifically for OpenClaw, to help make it “enterprise ready”.
“It finds OpenClaw, downloads it. It builds an AI agent for you,” Huang said.
In Automotive, Huang gave details of the previously announced partnership UberAnnouncing the Ride-Hail service will launch a fleet powered by Nvidia’s Drive AV software in 28 cities on four continents by 2028, starting with Los Angeles and San Francisco next year.
Nissan, BYD, Geely, Isuzu and Hyundai are building Level 4 autonomous vehicles in Nvidia’s Drive Hyperion program, Huang announced. Isuzu and China’s Tier IV are also building autonomous buses using the platform, with help from Nvidia’s AGX Thor robotic system chip.
— CNBC’s Jordan Novett contributed to this report.
Watch: Inside Nvidia’s Vera Rubin AI system
