General Intuition, a startup that builds AI world models from video game footage, has raised $220 million at a post-money valuation of $6.2 billion, SiliconANGLE reported Tuesday. The round was backed by Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst.

What Happened

The raise comes about three months after General Intuition's previous round. In June the company raised $320 million at a $2.3 billion valuation, according to GamesBeat, which puts its total funding to date above $650 million. The new price is roughly 2.7 times the June figure.

The company spun out of Medal B.V. about a year ago. Medal offers free tools that let players record and share clips of their games, and General Intuition trains its models on that footage. GamesBeat, which identified Pim de Witte as chief executive, reported that Medal is on track for about 3 billion uploads a year. The company told GamesBeat that leading robotics and world models are trained on less than 1% of the action data it is able to use.

RelatedWaymo Lines Up $5 Billion Loan From Blackstone, PIMCO and Other Lenders in First Debt Raise

SiliconANGLE said the new capital will mainly go toward hiring more AI researchers.

How MIRA Works

General Intuition's flagship model, MIRA, was released in June 2026. It has 5.6 billion parameters and can generate video at 20 frames per second at a resolution of 720x576 on a single Nvidia B200 graphics processor, according to SiliconANGLE. The model uses latent diffusion, working on compressed representations of video rather than raw pixels.

Its headline capability is stability over long runs. Many generative video systems drift or break down as sequences get longer; General Intuition says MIRA can keep rendering indefinitely without diverging.

The limits are significant. SiliconANGLE described MIRA as a research demonstration rather than a product, and it currently generates synthetic footage from only one video game. A commercial version is being tested with a small number of customers in robotics, simulation and entertainment, and the company opened a waitlist alongside the funding announcement.

Why It Matters

World models generate simulated environments that AI systems can learn from, which makes them a candidate tool for training robots and software agents without collecting as much expensive real-world data. General Intuition's pitch rests on a data source few rivals have: large volumes of gameplay video paired with the controller inputs that produced it. GamesBeat compared the approach to the way Tesla trained its Full Self-Driving software on camera footage linked to driver actions.

That pairing matters because a model that learns which actions lead to which visual outcomes can, in principle, be used to predict and plan, not just to produce video.

Context

Investor interest in world models has accelerated. On Monday, AMD agreed to buy Fei-Fei Li's World Labs for $8.2 billion, giving the chipmaker its own simulation models for robot training, an area where Nvidia already offers its Cosmos models. General Intuition's round shows private investors are also willing to pay rising prices for companies in the field before they have commercial products at scale.

What to Watch

The key test is whether MIRA can move beyond single-game demonstrations into environments varied enough to be useful for robotics customers. Results from the limited commercial trials, the pace of researcher hiring and any named robotics partners will indicate whether the valuation reflects more than access to Medal's footage.