The company that built the picks and shovels for the AI gold rush is now exporting the miners themselves.
The Summary
- Nvidia robotics veterans are leaving to launch startups, spinning out companies that build on the physical AI infrastructure Nvidia created
- Former VP of AI Research Sanja Fidler founded Veeda AI in July 2024 to build simulated training worlds for robots, directly extending her Nvidia research
- Nvidia invests in select alumni startups and maintains close partnerships, including biweekly meetings and early tool access, treating former employees as external beta testers
- The exodus signals that Nvidia's robotics platform has matured enough that the next value layer is application companies, not infrastructure
The Signal
Nvidia spent years building the rails for physical AI. Now the engineers who laid those rails are building the trains. Sanja Fidler, Nvidia's former VP of AI research, left in July to found Veeda AI with other Nvidia alumni. The company builds simulated environments for training robots. That is not a pivot. That is the exact research Fidler ran inside Nvidia, now packaged as a commercial product.
This is not brain drain. This is ecosystem multiplication. Nvidia is not losing talent. It is exporting it to companies that will buy more Nvidia chips and use more Nvidia software. Joel Jang, another former Nvidia researcher, founded Dream Labs in the same simulation space. Flexion Robotics and Dyna Robotics, both founded by Nvidia alums, have Nvidia as an investor.
"We're still doing the same thing we were doing before. We were the beta testers — and now again, just from the outside." — Nikita Rudin, Flexion Robotics CEO
Flexion holds biweekly meetings with Nvidia. They get early access to tools for testing. Nvidia treats these startups like an extended R&D lab that it does not have to pay salaries for. The startups get infrastructure, capital, and validation. Nvidia gets a growing market for its hardware and software. Everyone wins, as long as the applications layer takes off.
The timing matters. Physical AI is hitting the deployment phase. Simulation tools, warehouse robotics software, and task-specific robots for businesses are all transitioning from research projects to commercial products. Nvidia built the foundation. Its alumni are building the first floor. The question is whether venture capital will keep funding the build-out or if the market stalls before these companies prove revenue models.
Key dynamics in play:
- Nvidia's robotics platform is mature enough that application-layer companies are viable
- Venture capital is flooding into physical AI, funding the alumni founder wave
- Nvidia maintains close ties and invests selectively, ensuring alumni startups stay in its ecosystem
Amulya Vishwanath, who left Nvidia in 2023 to start Techable Ventures, a physical AI-focused VC firm, sees this as a predictable pattern. The infrastructure is built. The talent knows how to use it. The capital is available. The next wave is always founders who worked on the platform spinning out to build on top of it.
The Implication
Watch where these founders cluster. Simulation for robot training is hot right now because it solves a hard problem: you cannot train robots in the real world at scale without breaking expensive hardware. If three or four well-funded startups are attacking that problem, one of them will likely win and become a major Nvidia customer. Same logic applies to warehouse deployment software and task-specific robotics.
For anyone building in robotics, the takeaway is clear. Nvidia is not just a chip vendor. It is a platform play. If you are building on Nvidia's stack, you get access, early tools, and potentially capital. If you are building against it, you are fighting uphill. The alumni network is now a competitive moat.