Nvidia's Trillion-Dollar Rise Fueled by Modest Research Lab

In 2009, when Bill Dally joined Nvidia's research division, it was a small team of about a dozen people focused primarily on ray tracing for computer graphics.

That modest research group has since grown to over 400 employees, playing a pivotal role in Nvidia's evolution from a 1990s gaming GPU startup into a $4 trillion company at the heart of the artificial intelligence revolution.

The lab's focus has now shifted to developing the foundational technology for robotics and AI, with some of its work already reaching the market. On Monday, the company introduced a new suite of world AI models, libraries, and developer infrastructure for robotics.

Dally, now Nvidia's chief scientist, first consulted for the company in 2003 while at Stanford. Years later, as he prepared to step down as chair of Stanford's computer science department for a sabbatical, Nvidia had other plans for him.

David Kirk, then head of research, and CEO Jensen Huang believed a permanent role at the lab was a better fit. Dally told TechCrunch the pair mounted a "full-court press" to convince him to join, and they eventually succeeded.

"It turned out to be a perfect match for my interests and skills," Dally said. "Everyone searches for where they can make their greatest contribution. For me, that place is undoubtedly Nvidia."

When Dally assumed leadership of the lab in 2009, expansion was the immediate priority. The team quickly branched out from ray tracing into new areas like circuit design and VLSI (very-large-scale integration), the process of embedding millions of transistors onto a single chip.

The research lab has been growing ever since.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital, Elad Gil — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $600+ before prices rise.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $675 before prices rise.

"We aim to identify what will have the greatest positive impact for the company," Dally said. "We constantly explore exciting new areas, but with some, it's difficult to predict if we'll achieve wild success, even with excellent work."

For a significant period, that focus was on building superior GPUs for artificial intelligence. Nvidia anticipated the AI boom early, beginning to develop AI GPU concepts in 2010—over a decade before the current frenzy.

"We recognized this was revolutionary and would change the world," Dally recalled. "We had to double down. Jensen believed me when I told him. We started specializing our GPUs for AI, developing extensive software, and engaging with global researchers long before its relevance was obvious."

Physical AI focus

Now, with a commanding lead in the AI GPU market, Nvidia is exploring new demand areas beyond AI data centers, leading them to physical AI and robotics.

"Robots will eventually become a massive global industry, and we aim to provide the brains for all of them," Dally said. "To achieve that, we need to develop the core technologies now."

This is where Sanja Fidler, Nvidia's vice president of AI research, enters the picture. Fidler joined the research lab in 2018, bringing expertise from her work on robot simulation models with a student team at MIT. When she discussed her research with Huang at a reception, he was immediately intrigued.

"I couldn't resist joining," Fidler told TechCrunch. "It was a perfect topic and cultural fit. Jensen's invitation was personal—'come work with me,' not just for the company."

She joined Nvidia and established a Toronto research lab focused on Omniverse, a platform dedicated to building simulations for physical AI.

Fidler stated the first major challenge was sourcing the necessary 3D data. This involved finding sufficient image volumes and creating the technology to convert these images into usable 3D models for simulators.

"We invested in a technology called differentiable rendering, which makes rendering compatible with AI," Fidler explained. "Traditional rendering goes from 3D to image. We needed to invert that process."

World models

Omniverse launched its first image-to-3D model, GANverse3D, in 2021. The team then tackled the same process for video. Using footage from robots and self-driving cars, they created 3D models and simulations via the NeRF (Neural Reconstruction Engine), announced in 2022.

Fidler noted these technologies form the foundation of the company's Cosmos family of world AI models, unveiled at CES in January.

The lab's current priority is increasing the speed of these models. For robots and simulations, response times must be real-time, Fidler emphasized, and for robotics, they aim for even faster reaction capabilities.

"A robot doesn't need to perceive the world at the same speed we do," Fidler said. "It can process information 100 times faster. Significantly accelerating these models will make them immensely useful for robotic and physical AI applications."

The company is advancing toward this goal. At the SIGGRAPH computer graphics conference on Monday, Nvidia announced a new fleet of world AI models for generating synthetic data to train robots, along with new libraries and infrastructure software for robotics developers.

Despite the progress—and the current hype around robots, particularly humanoids—the Nvidia research team remains pragmatic.

Both Dally and Fidler estimate the industry is still several years away from practical home humanoid robots, with Fidler comparing the timeline to that of autonomous vehicles.

"We're making tremendous progress, and AI has been the key enabler," Dally said. "It started with visual AI for robot perception, and now generative AI is proving immensely valuable for task planning, motion planning, and manipulation. As we solve each sub-problem and our training datasets expand, these robots will continue to evolve."

We're always working to improve. Sharing your perspective and feedback on TechCrunch's coverage and events can help us evolve. Fill out this survey to let us know how we're doing—you'll also have a chance to win a prize!