When I walked into General Intuition's New York R&D office on Thursday, CEO Pim de Witte wasn't interested in small talk. The 31-year-old Dutch entrepreneur pointed me toward a monitor where something resembling Fortnite was running. It wasn't someone playing โ€” it was an AI agent that had been grinding the game for 100 hours straight without a break. "The same brain powering the agent playing the game is powering the robot," de Witte told me as a large quadrupedal machine clanked across the floor behind us, navigated by that exact same model. That convergence of gaming data and physical robotics just secured General Intuition $320 million in fresh funding at a $2.3 billion valuation โ€” bringing total disclosed capital to $454 million after its October launch round of $134 million.

The Action Label Advantage

Most AI labs trying to build world models are working backward from video footage, attempting to infer what actions caused what outcomes. General Intuition takes a different approach: they already know the answers. De Witte's other company, Medal โ€” which lets gamers upload and share clips โ€” provided hundreds of millions of hours of gameplay data. But here's the critical piece that separates their method from competitors: those clips contain action labels. Every button press, every mouse movement, every exact moment a player decided to jump or shoot or build was captured and labeled. "We view this as just the next stage of future pre-training," de Witte explained. "We have a single model that can respond to Fortnite information on the screen and take action, but also to real-world dynamics in a way that an LLM could never." During my demo with their world model โ€” a frame-by-frame simulated environment rather than a traditional game engine โ€” I walked straight into walls repeatedly. The agent didn't let me clip through them like I've seen happen elsewhere. It had somehow internalized from millions of hours of gameplay that walls are obstacles, ladders are for climbing, and shadows shift as the sun moves.

Why Khosla Went All-In

The round was led by Vinod Khosla, who doesn't throw his weight around lightly. He told me over the phone that he sees General Intuition as a generational bet, not an acquisition target โ€” and that's by design. "If you look at LLMs, when reasoning emerged, it was a quantum leap," Khosla said. "In world models, I think the quantum leap is the emergence of intuition in the AI, a human intuition-like capability. The human action data and reaction data you have in games is the key part to the emergence of intuition." Other investors signing on include Jeff Bezos, Eric Schmidt, General Catalyst, Nico Rosberg, and researchers from Google DeepMind and MIT. De Witte's team isn't based in Silicon Valley โ€” much of it is European by design. He spent three years working humanitarian operations including with Doctors Without Borders before pivoting to AI. That background informs the company's hard line: no agents will be employed to harm humans. "We don't want to be an escalatory part of the system," de Witte said, noting he'd be fine with search and rescue deployments but drew a clear boundary on lethal autonomy applications.

Building the Data Flywheel

General Intuition isn't trying to become a self-driving car company or a robotics manufacturer. The goal is ecosystem infrastructure โ€” think Anthropic or OpenAI positioning, where they provide the foundational model and let others build on top of it. The company already has customers in gaming, simulation, and robotics testing. Josh Duplantis, a data analyst at the company, showed me how they fine-tuned their quadruped model using just eight minutes of real-world robotics footage collected on an actual street โ€” not in the office where the bot was currently bumping into furniture like a toddler learning spatial awareness. "It works on anything that you can control using a game controller or a keyboard mouse," de Witte noted, mentioning they've also tested drones and driving games beyond quadrupeds. The API becomes broadly available by end of summer, which will let more customers stress-test the simulation-to-reality transfer at scale โ€” the key question nobody has fully answered yet in this space.

Gaming's Unexpected Workforce Pipeline

Here's where it gets interesting for the gaming community: General Intuition recently launched Nerve, a jobs marketplace designed to let gamers earn money using their existing hardware setups. Workers start with data labeling tasks and can eventually move toward robot teleoperation and other roles as the platform matures. De Witte made $1.5 million building RuneScape servers as a teenager โ€” he's acutely aware that Medal's user base represents precisely the generation most exposed to AI-driven job displacement, and he wants them positioned to benefit rather than get left behind. "We're gonna make it 10 times easier for the next person to build a self-driving car company," de Witte said about their API strategy. The bulk of this new funding goes toward compute capacity via a CoreWeave partnership and pre-training the next model version.

Key Takeaways

  • General Intuition raised $320M at $2.3B valuation, backed by Khosla Ventures, Bezos, Schmidt, and Google DeepMind researchers
  • Their advantage: action labels from Medal's video game clips reveal exactly what buttons players pressed when โ€” not just visual outcomes
  • The same model that plays Fortnite for 100 hours straight also runs a quadruped robot fine-tuned with only 8 minutes of real-world data
  • No military applications allowed by design; company is targeting search and rescue, simulation testing, and ecosystem infrastructure instead

The Bottom Line

This is either the most pragmatic use of existing gaming data anyone's figured out yet, or an audacious bet that simulation-to-reality transfer actually scales โ€” but with Khosla, Bezos, and Schmidt all writing checks simultaneously, someone's convinced the math works. Either way, gamers grinding clips on Medal just became inadvertent infrastructure for the next generation of robotics.