A $90M seed says robotics' scarce input is now the environment, not the robot

Veeda AI, a startup led by former Nvidia researcher Sanja Fidler, has emerged with a $90 million seed round co-led by Khosla Ventures and Radical Ventures, three months after the company was founded ^[1][2][3]. SiliconANGLE, citing The Logic's first report, calls it one of the largest seed financings ever raised by a Canadian startup ^[4][5]. The consequence for anyone building embodied systems is a repricing of inputs: the money is going to the environment, not the machine.

The company, formally Veeda Innovation Inc., builds multimodal world models meant to simulate the physical world, with the stated goal of infinitely scalable simulated environments for training embodied agents through repeated interaction ^[6][7]. Fidler's argument, made in a LinkedIn post, is a scaling argument rather than a capability one: robotic hardware does not scale the way compute does, real-world mistakes can be dangerous, and real-world experience cannot be parallelized ^[8]. "To scale interactive learning, robots will need to learn in a simulated reality," she wrote ^[9]. Veeda describes what it is building as a kind of "Matrix" for physical AI, and Fidler says the conviction is that simulated reality "will become the critical infrastructure layer for all areas of robotics" ^[10][11].

That is a bet on a ratio. If experience is the binding constraint, then a fleet of physical robots is the worst possible way to buy it, and GPU hours plus generated worlds are the cheapest. The claim that operators cannot check from this announcement is the one that matters most: nothing in the reporting quantifies how well policies trained in Veeda's environments transfer to hardware, and no product, customer, benchmark or pricing unit is disclosed ^[12]. What is priced here is provenance. Fidler joined Nvidia in 2018 to help establish its Toronto research unit, which became the company's Spatial Intelligence Lab, where her team built software simulating how autonomous vehicles and humanoid agents interact in the real world ^[13][14]. She and co-founder Huan Ling later helped develop Nvidia's first world models for physical AI developers; Zan Gojcic is the third co-founder ^[15][16].

The cap table is the more instructive artifact. Khosla and Radical co-led the seed round of Jeff Dean's startup Discovery Loop earlier this month ^[17]. Khosla was a lead investor in autonomous trucking company Waabi's $750 million round in January, with Radical participating; Waabi builds world models for training vehicles to drive autonomously ^[18][19]. Radical has also backed Fei-Fei Li's World Labs, which raised $500 million in January, and Decart.ai, which raised $300 million in May ^[20][21]. Counting Veeda, that is roughly $1.64 billion across four world-model or spatial-intelligence companies named in a single article ^[22]. Two firms are building a portfolio on the premise that simulation is a layer, not a feature.

Three things to watch. First, whether these companies sell simulation to robot makers or get absorbed by them, since the buyers are also the parties best placed to generate their own data. Second, the unit of sale: a simulation platform priced per environment, per hour or per frame implies very different margins, and none has been named ^[12]. Third, how the four bets relate to each other, given that Waabi's driving-focused world models and Veeda's general physical-AI simulation share an investor and, in part, a thesis ^[19][7].