Via nvidia.com
The small robotics AI startup's struggles reveal a funding gap that could slow the next wave of artificial intelligence beyond the screen.
Physical AI, the branch of artificial intelligence that controls robots and autonomous vehicles in the real world, has a money problem. Integral AI, a startup founded by former Google researchers Jad Tarifi and Nima Asgharbeygi, is learning that lesson the hard way.
The company, which built what it described as foundational world models for robotics and self-driving systems, raised roughly $4.7M to $5.5M in seed funding from backers including SoftBank’s Deepcore and Samsung Next. That sounds like a decent start until you realize the scale of capital required to make robots reliably learn new tasks in unpredictable physical environments.
The gap between hype and hardware #
Integral AI’s pitch was ambitious. In December 2025, the company announced what it called the “world’s first AGI-capable model,” one that would enable robotic systems to autonomously learn new skills without labeled data. The team of roughly 15 employees had been working with partners like Denso Corp. since 2021, focused on industrial robot skill acquisition. Toyota and Sony were among the major Japanese manufacturers the company was engaging with.
As of March 2026, Integral AI was seeking approximately $10M in its next funding round to accelerate development.
Training a large language model requires vast quantities of text scraped from the internet, which is expensive but at least the data exists. Training a robot to navigate a warehouse or assemble components requires physical interaction data, often collected painstakingly through real-world trials or elaborate simulations. Every data point costs more. Every failure mode carries higher stakes.
A two-tier funding market #
Physical AI compounds this problem. Unlike software-only AI companies that can demonstrate progress with a chatbot demo or a benchmark score, robotics startups need to show working hardware, real-world deployments, and safety records. The burn rate is higher. And the path to revenue is murkier, especially for companies building foundational models rather than end-to-end products.
Integral AI had some of the ingredients that should attract capital: pedigreed founders from Google, partnerships with major industrial players, and a technically differentiated approach to autonomous learning. The fact that even this combination proved insufficient to smoothly secure follow-on funding tells you something about how steep the hill is for physical AI ventures.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our