Robotics Startup General Intuition Raises $320 Million for Foundation Model Trained on Video Games

Startup General Intuition has raised $320 million at a $2.3 billion valuation, betting that robotics is one step away from its own ChatGPT moment. Instead of collecting expensive, specialized data from robot sensors, the company trains its model on millions of hours of video game footage, then fine-tunes it to control physical machines using just a few minutes of real-world data.

The news was first reported by TechCrunch on July 8, 2026, based on an interview with the company's founder and CEO, Pim de Witte. General Intuition claims to have solved one of robotics' most costly problems: the lack of training data describing physical action in the real world.

Video games as training data

Rather than building fleets of robots to collect data in warehouses or factories, General Intuition trains its model on video game footage, recording player actions and controller inputs. This lets the model learn general patterns of movement, planning, and reacting to changing surroundings before it ever encounters real hardware.

The result is a model that can play games for hours without losing consistency. But the key test came when that knowledge was transferred to the physical world: after fine-tuning on just eight minutes of real-world data collected from a four-legged robot, the model took over control of it.

From joystick to robot

Pim de Witte argues that the robotics industry is mistakenly repeating an early approach to artificial intelligence, in which companies poured resources into narrow, specialized datasets for individual tasks. In his view, the real value lies not in the dataset itself but in the model's ability to generalize, transferring knowledge gained in one environment to entirely different physical tasks.

General Intuition does not plan to build its own robots or compete with hardware makers. The company wants to serve as a foundational layer on which other teams build specific products, from warehouse robots to autonomous vehicles, without having to train their own models from scratch.

A bet on foundations

The $320 million round and $2.3 billion valuation show that investors are increasingly betting on foundation models for robotics as a category of its own, separate from hardware manufacturers. Vinod Khosla, who led the round, has spent years betting on infrastructure layers of artificial intelligence before they become obvious to the rest of the market.

It's part of a broader trend from the first half of 2026, in which venture capital is shifting more heavily toward AI operating in the physical world, alongside foundation models for financial markets and agentic systems for narrow, regulated industries.

The generalization of the model itself is the product - Pim de Witte, founder and CEO of General Intuition

For the Polish market, where companies like Nomagic are developing their own models for controlling warehouse robots, General Intuition's approach is both a warning sign and an opportunity. If foundational motion models become available as a ready-made technology layer, smaller robotics teams will be able to build specific products faster and more cheaply, instead of spending months collecting their own sensor training data.

The biggest open question is whether motion patterns learned from video games really do transfer to complex industrial tasks to the degree the company claims, or whether the four-legged robot demo is, for now, an impressive showcase rather than a production-ready solution. The answer will only become clear once General Intuition's customers start deploying the technology in real work environments.

Sources: This startup thinks robotics is about to have its ChatGPT moment (techcrunch.com)