A startup just raised $320 million on a contrarian idea: the path to smarter AI runs through video games, not the open internet. That’s the pitch from General Intuition, a Bezos-backed, New York company now valued at $2.3 billion, and CEO Pim de Witte laid out the thesis on TechCrunch’s Equity podcast, as reported by TechCrunch AI. His argument cuts against the grain of how most frontier labs think about training data.
Here’s the core claim. Large language models like ChatGPT and Claude are excellent with text, but they struggle to understand how objects actually move through space and time. De Witte calls that gap a dealbreaker for artificial general intelligence. Gaming footage, he argues, captures exactly what text leaves out: motion, physics, cause and effect, and how an agent navigates a world.
What the company is building
General Intuition is chasing what researchers call world models. Instead of predicting the next word, a world model learns to predict the next state of an environment: what happens when something moves, collides, or reacts. Train that on enough gameplay and you get a system with intuition about physical space, the kind of grounding text alone can’t provide.
The company spun out of Medal TV, a gaming clip platform, which gives it a built-in pipeline of gameplay data most competitors would have to build from scratch. That data advantage is a big part of the bet.
The round pulls in serious names, according to TechCrunch AI:
- Coatue leading the investment
- Eric Schmidt, former Google CEO
- Researchers from MIT and Google DeepMind
- Jeff Bezos among the backers
When DeepMind researchers and Schmidt put money behind a spinout from a gaming clip site, that’s a signal worth watching.
Why this matters
The AI industry has largely operated on one assumption: scale up text and code from the internet, and capability follows. That approach is hitting real limits. The web is close to fully scraped, high-quality text is getting scarce, and text-trained models still fumble basic physical reasoning. Ask a chatbot to reason about a stack of blocks or a robot arm reaching for a cup, and the cracks show.
General Intuition is part of a broader shift toward physical AI, systems meant to operate in the real world rather than just chat about it. Robotics, autonomous machines, and embodied agents all need the spatial understanding that de Witte says gaming data can teach. What stands out here is the framing: gaming isn’t a niche use case, it’s positioned as better training data than the internet for the skills that generalize.
This is significant because it reframes where the next capability jump might come from. If world models trained on gameplay deliver real spatial reasoning, the companies sitting on large stores of video and interaction data suddenly look a lot more valuable than the ones racing to scrape the last corners of the web.
The uncomfortable part
De Witte didn’t dodge the harder question. On the podcast he addressed where the ethical red lines sit when models built on gaming data could end up in defense applications. A system that understands how agents move and act in a simulated world has obvious dual-use potential. It’s a live tension for any company building physical AI, and it’s notable that he’s raising it out loud rather than waiting to be asked.
What to watch next
A few things worth tracking:
- Proof it works. A $2.3 billion valuation is a bet on a thesis, not a shipped product. The real test is whether gaming-trained world models outperform text-trained systems on physical reasoning.
- The data land grab. If this thesis holds, expect more companies to start treating video and interaction data as strategic assets.
- Robotics crossover. World models with spatial intuition are exactly what robotics teams have been missing. Watch for partnerships or applied demos.
- The defense question. How General Intuition draws its ethical lines will shape how seriously the industry takes its governance.
Whether gaming data really beats the internet is still an open question. But with this much capital and this caliber of backer behind the idea, it’s one the whole field will be forced to reckon with. Full details are in the Equity podcast episode at the original source.