General Intuition Data Loop Advantage

The moat here is not just having game video, it is owning a live machine that keeps turning user behavior into better training data, better models, and more cash to buy more compute. Medal gives General Intuition a constant stream of clips tied to actual player inputs, which is much harder to assemble than passive video because it needs an existing consumer product, large active usage, and infrastructure to ingest and train on massive action labeled datasets.

• A new entrant cannot simply buy this dataset off the shelf. Medal is already producing billions of clips per year at consumer scale, and CoreWeave describes General Intuition training on more than 1B hours of Medal video, which means the data asset is large, fresh, and operationalized into production training workflows.
  1 sacra 3 medal 4 techcrunch
• The data is more useful than ordinary scraped video because each clip carries a record of what buttons were pressed and what happened next. That lets models learn cause and effect, like turning left, boosting, colliding, or scoring, instead of only learning what game footage looks like frame by frame.
  4 techcrunch
• Competitors face a harder starting point depending on their strategy. Horizontal model labs can match compute but not this proprietary action stream, while robot makers like Figure and Apptronik have their own closed loop only after deploying hardware fleets, which is slower and more capital intensive than harvesting gameplay behavior from an existing network.
  4 techcrunch

This points toward a market where the strongest model providers are the ones that already control a native data engine. If General Intuition keeps converting Medal scale into better pretraining, the company can widen the gap before rivals build equivalent consumer funnels or accumulate enough real world robot interaction data to challenge it.