Every few months a new data vertical gets anointed as the thing LLMs supposedly can’t live without, and this cycle it’s gaming data’s turn. General Intuition’s pitch, backed by Bezos Expeditions, is that text-trained models like ChatGPT and Claude lack a working model of how objects move through space and time — precisely the kind of signal that millions of hours of player inputs, camera movement, and physics interactions could supply cheaply and at scale.
If that thesis holds, it reprices an entire category of data that studios and platforms have mostly treated as exhaust. Telemetry, replay logs, and control-input streams from shooters, racing games, and open-world titles suddenly look less like analytics fodder and more like licensable training corpora that could compete with the synthetic-simulation data Nvidia, World Labs, and robotics labs are already generating. Expect gaming publishers to start asking the same question mobile apps and forums asked two years ago: who owns this, and what should a frontier lab pay for it?
The moment gameplay logs get treated as a distinct AGI-training asset class, every studio with a large install base becomes a potential data seller — whether or not it wants the job.
The harder question is whether gaming data actually closes the spatial-reasoning gap or just adds another narrow domain to the pile, the same skepticism that’s dogged robotics and video-pretraining claims before it. Watch for whether General Intuition announces actual licensing terms with studios or publishers, and whether any frontier lab — OpenAI, Google DeepMind, or a robotics-focused buyer — puts a real check behind the idea that your K/D ratio is secretly AGI training data.
When it comes to achieving artificial general intelligence (AGI), large language models just don't have what it takes. Models like ChatGPT and Claude are great at text, but they're less skilled at understanding how things actually move through space and time — an essential skill for producing intelligence that generalizes. That gap, it turns out, might be filled by gaming data.