Ex-Nvidia AI research VP Fidler raises $90M+ seed for world-model startup
A massive seed round just landed in the world-model space. Veeda, a new startup building world models and led by Sanja Fidler — formerly Nvidia's VP of AI Research — has raised over $90 million in seed funding from Khosla Ventures and Radical Ventures, according to The Logic.
World models — systems that learn to simulate how the physical world works — have become one of the hottest frontiers in AI since NVIDIA open-sourced Cosmos 3 earlier this year. The idea is straightforward but ambitious: instead of training AI on text and images alone, you teach it the underlying physics and dynamics of reality, letting it predict outcomes, plan actions, and reason about embodied scenarios without real-world trial and error. Fidler's team at Veeda is now one of the best-funded attempts to push that vision into production.
Fidler brings serious credentials to the bet. At Nvidia she led AI research efforts spanning computer vision, neural rendering, and autonomous driving perception — areas where world models could unlock the next generation of capabilities. Before Nvidia she spent years at the University of Toronto and SRI International, building a reputation as one of the leading researchers in visual AI. Her move to start Veeda signals that world models have crossed the threshold from academic curiosity to commercial priority.
The investors reinforce that signal. Khosla Ventures, led by Vinod Khosla, has been one of the most aggressive backers of frontier AI companies. Radical Ventures, a Toronto-based fund with deep ties to the Canadian AI ecosystem and connections to Geoffrey Hinton, has specifically focused on backing transformative AI research. A $90 million seed round is exceptionally large even by today's inflated standards — it suggests Veeda already has a working prototype or early results strong enough to command venture-scale conviction before a Series A.
The timing matters. NVIDIA's Cosmos 3 launch demonstrated that open-source world models can run at real-time speeds on consumer hardware, validating the technical approach. But Cosmos is a general-purpose platform — Veeda appears to be going after specific vertical applications where world models deliver outsized value, likely in robotics, autonomous systems, or simulation-heavy industries. The world-model space is crowded quickly, and a seed round of this size is Veeda's way of buying a head start.
Harness, the DevOps and software delivery platform valued at $5.5 billion, shipped a new suite of AI agents today that find software vulnerabilities and automatically write patches. The agents run a deterministic static scanner, apply an AI layer to strip false positives, then triage findings for exploitability before drafting and validating fixes. A zero-day agent watches newly disclosed flaws around the clock and flags affected systems across a customer environment, often with a validated fix ready within minutes.
The launch is notable for what it says about the AI security arms race: Harness claims frontier models now surface roughly ten times more vulnerabilities than conventional scanners, but most security teams lack the capacity to process that flood of findings. By pairing detection with automated remediation, Harness is betting that the solution to AI-amplified threats is equally AI-amplified defense. The agents draw on Anthropic's Claude Mythos Preview model, which Anthropic has kept restricted due to its effectiveness at finding and chaining software vulnerabilities.
What to watch: Whether Veeda publishes technical results or benchmarks that distinguish its world models from Cosmos 3 and other open-source alternatives — early demos will determine if this is a platform play or a vertical product.
Does a $90 million seed round signal that world models are about to go mainstream, or is the space still too early for this kind of capital commitment? Tell us in the comments.
Sources: The Logic · Techmeme · SiliconANGLE
