General Intuition $133.7M: Gaming Data Powers Spatial-Temporal AI Agent Reasoning
Medal’s gaming clip platform generated 2 billion videos per year from 10 million active users — now its AI spinoff General Intuition has raised $133.7 million in seed funding to solve what CEO Pim de Witte calls the fundamental limitation of text-trained AI: spatial-temporal reasoning.
The massive seed round, led by Khosla Ventures and General Catalyst with participation from Raine, represents one of the largest AI infrastructure investments of 2025. More significantly, it validates a contrarian bet that gaming data holds the key to building AI agents capable of navigating and reasoning about the physical world — capabilities that current language models fundamentally lack.
The Spatial Reasoning Bottleneck
While the AI industry has focused on scaling language models, General Intuition argues that true agent intelligence requires understanding how objects and entities move through space and time. “As humans, we create text to describe what’s going on in our world, but in doing so, you lose a lot of information,” de Witte explained. “You lose general intuition around spatial-temporal reasoning.”
This bottleneck becomes critical when deploying AI agents in real-world scenarios requiring navigation, prediction, and coordination. Current text-trained models excel at language tasks but struggle with fundamental spatial concepts like object persistence, movement trajectories, and environmental interaction — precisely the skills needed for autonomous operations.
Gaming data offers a unique solution: billions of first-person navigation sequences where humans demonstrate spatial reasoning in complex, dynamic environments across thousands of different game worlds.
Gaming Data Moat Architecture
General Intuition’s approach leverages Medal’s massive dataset advantage: 2 billion gaming clips annually from diverse genres, each capturing human spatial decision-making in real-time. Unlike passive video platforms, Medal’s clips represent active gameplay moments — successful strategies, failure cases, and edge scenarios that provide rich training data for agent behavior.
“When you play video games, you essentially transfer your perception, usually through a first-person view of the camera, to different environments,” de Witte noted. Gamers upload both spectacular successes and notable failures, creating the exact edge cases needed for robust agent training.
The technical architecture focuses purely on visual input — agents see only what human players see and navigate using controller inputs. This design enables direct transfer to physical systems like robotic arms, drones, and autonomous vehicles, which often use gaming controllers for human operation.
Enterprise Validation and Market Evidence
OpenAI’s reported $500 million acquisition attempt for Medal late last year signals major AI labs recognize gaming data’s strategic value. General Intuition’s ability to secure $133.7 million in seed funding — among the largest AI infrastructure rounds — demonstrates investor confidence in spatial reasoning as a critical capability gap.
The company has already demonstrated transfer learning beyond gaming environments: their models can understand and navigate spaces they weren’t explicitly trained on, correctly predicting actions in novel environments through spatial-temporal reasoning skills developed on gaming data.
Initial commercial applications target gaming NPCs that scale difficulty dynamically and search-and-rescue drones requiring GPS-free navigation in unfamiliar terrain — both scenarios where spatial reasoning under uncertainty becomes mission-critical.
Infrastructure Implications
General Intuition’s success highlights the emerging “data moat” era in AI infrastructure, where unique datasets become more valuable than pure computational scaling. Gaming data represents a particularly rich source because it captures human spatial intelligence across diverse environments, weather conditions, and obstacle patterns.
Unlike text data, which compresses three-dimensional reality into linear descriptions, gaming footage preserves the full spatial-temporal context needed for real-world agent deployment. This preservation of dimensional information could prove essential as AI systems move beyond chat interfaces toward autonomous physical operations.
The gaming industry’s existing infrastructure for capturing, storing, and processing massive video streams also provides a ready-made platform for spatial reasoning data collection — avoiding the complex logistics of building physical world data collection systems.
Looking Forward: World Model Competition
General Intuition’s $133.7 million positions it alongside other world model companies like DeepMind’s Genie and World Labs’ Marble, but with a distinctly different commercial strategy. Rather than selling world models directly, General Intuition focuses on gaming applications and physical world navigation to avoid content creation copyright concerns.
The next 12 months will test whether gaming-trained spatial reasoning can successfully transfer to real-world robotic systems and whether deterministic gaming environments provide sufficient edge case coverage for unpredictable physical scenarios. Success could establish gaming data as a new infrastructure category for agent training.
Early evidence suggests spatial-temporal reasoning represents a fundamental capability gap that text-scaling approaches cannot address — making General Intuition’s gaming data approach a potential infrastructure requirement for any AI system requiring real-world navigation and coordination capabilities.
General Intuition’s approach demonstrates how specialized data sources can address fundamental AI capability gaps that pure scale cannot solve. As AI agents move beyond text interaction toward physical world operations, spatial-temporal reasoning infrastructure becomes increasingly mission-critical.
Overclock enables teams to deploy these spatial reasoning capabilities through orchestrated agent workflows, bridging the gap between research breakthroughs and production enterprise applications requiring real-world navigation intelligence.