A large language model can draft your yearly business plan or write a sonnet, but it cannot reliably help a robot pick up a cup from your table. This gap between cognitive and physical ability represents one of the biggest bottlenecks in technology today, a challenge rooted not in algorithms, but in data.
While language models feasted on trillions of words scraped from the internet, a comparable dataset for physical interaction simply does not exist. There is no public library of how a hand twists a doorknob or how a foot navigates a cluttered floor.
Into this void steps Veeda AI, a Toronto-based, Slovenian co-founded startup armed with one of the largest seed rounds in Canadian history. Founded by a trio of elite researchers who recently departed chip giant Nvidia, the company has secured over $90 million (€83.7M), led by Khosla Ventures, to tackle the physical world not by interacting with it, but by perfectly simulating it.
Their mission is to build the foundational infrastructure for physical AI, creating infinitely scalable virtual worlds where robots can learn, fail, and master tasks millions of times over before making their first move in reality.
“The next breakthrough is around the corner”
Veeda AI is led by CEO Sanja Fidler, a towering figure in Canadian AI who, until last month, served as Vice-President of AI Research at Nvidia, heading its Toronto-based Spatial Intelligence Lab. Fidler, who grew up in Slovenia and completed her doctorate at the University of Ljubljana, announced her departure after eight years, declaring in a LinkedIn post that “world models is where the next breakthrough lies, and it is around the corner.”
She is joined by co-founders Zan Gojcic, another Slovenian researcher and former Nvidia research director in Zurich who serves as CTO, and Huan Ling, a former colleague from the Spatial Intelligence Lab and Fidler’s past doctoral student, who is now Veeda's Chief Scientist. The team, which incorporated Veeda Innovation in early June, possesses a rare concentration of expertise in spatial intelligence, simulation, and generative AI.
Their vision quickly attracted heavyweight backers. The $90 million seed financing was co-led by Toronto’s Radical Ventures and Silicon Valley’s Khosla Ventures, placing the round in the 99th percentile of all seed deals in the industry. More than with just the capital, this investment will also serve as a strategic endorsement. As part of the deal, Radical partner Tomi Poutanen and Khosla partner Sven Strohband have joined Veeda’s board, signalling deep operational involvement from two of the most influential firms betting on the future of AI.
It is unsafe, expensive, and time-consuming
The central premise behind Veeda is a direct response to the intractable data problem in robotics. As Fidler explains, training a robot through physical trial and error is fundamentally unscalable. “The real world is simply not a practical training ground for robots to learn through trial and error. It is unsafe, expensive, and time-consuming,” she wrote. A mistake from a language model results in a bad sentence; a mistake from a multi-tonne industrial robot can be catastrophic.
Veeda’s solution is to build what it calls “simulated reality” powered by generative world models.
These are not simple video game environments. They are multimodal foundation models that learn the physics and mechanics of reality from colossal amounts of sensor data, including images, video, and text. The goal is to create a digital twin of the world so rich and responsive that an AI agent can learn within it, experiencing millions of scenarios and consequences in parallel. Fidler unabashedly calls it building the Matrix:
As envisioned in pop culture, this means building “the Matrix” for Physical AI, where, through a virtual embodiment, robot intelligence can interact with an entirely virtual world in a million possible ways, learning from its own mistakes.
The approach mirrors the evolution of language models, which only unlocked their full potential when they began training in interactive environments, learning from reinforcement and feedback rather than just passively observing internet text. Veeda is betting the same paradigm shift is necessary for embodied intelligence.
A well-worn playbook
For the founding team, this is not a new frontier.
At Nvidia, Fidler, Gojcic, and Ling were central to the company’s efforts in spatial intelligence and simulation.
Fidler’s Spatial Intelligence Lab developed software to simulate how autonomous vehicles and humanoid agents would interact with the real world. She and Ling were part of the team that created Nvidia’s flagship world model, Cosmos, a platform used by companies like Uber and Figure AI to build their own physical AI systems.
Their research has consistently pushed the boundaries of generating and editing three-dimensional digital content, from AI models that could reproduce a video game from gameplay footage to driving simulators built from real-world dashcam video.
Even though they went out of stealth just last week, this extensive background means Veeda isn’t starting from scratch. Veeda team is accelerating a research agenda they have pursued for the better part of a decade, now with the focus and funding of a dedicated startup.
The company is already hiring aggressively for its offices in Toronto, Zurich, Singapore, and Mountain View, seeking scientists and engineers to manage the massive chip clusters required to train its ambitious models.
The investors betting on the same ticket
The investor syndicate behind Veeda is perhaps as telling as the technology itself.
Radical Ventures and Khosla Ventures have developed a clear thesis around backing elite, research-led teams building foundational AI models. Earlier this year, the two firms co-led a founding round for Discovery Loop, an AI-for-science startup launched by former Google chief scientist Jeff Dean.
Their portfolios are heavily weighted towards Veeda’s specific domain. Khosla co-led the massive $750 million round for autonomous trucking company Waabi, which develops its own world models for self-driving (founded by Raquel Urtasun, another U of T professor). Radical was a participant in that deal and is also an investor in World Labs, a spatial intelligence company founded by its own scientific partner, Fei-Fei Li.
This pattern confirms important conviction of some of the most prominent investors: that the next wave of AI might be defined by models that understand and interact with the physical world, and that the talent emerging from top corporate labs and academic institutions is the best bet to build it.
While the first wave was defined by large language models, Veeda joins a cohort including Yann LeCun’s AMI Labs and General Intuition, all chasing the holy grail of physical intelligence.
With its $90 million war chest, Veeda is now positioned to compete for the two most critical resources in this race: elite research talent and the vast amounts of computing power needed to bring their simulated worlds to life. For roboticists everywhere, the message is clear. As Fidler concluded in her announcement, “Help us understand what you need, we’re building this for you.”
Member discussion