General Intuition raises $320M to train AGI on gaming data

General Intuition closed a $320 million funding round and is valued at $2.3 billion, the company announced on Jul 8, 2026. The Bezos-backed, New York-based startup says it will train world models on gaming data and build tools that connect gamers to labeling and teleoperations work.

What did General Intuition announce?

General Intuition raised $320 million and sits at a $2.3 billion valuation, with investors including Coatue, Eric Schmidt, and researchers at MIT and Google DeepMind. CEO Pim de Witte discussed the company on the Equity podcast, describing the startup’s origin as a spinout from gaming platform Medal TV and explaining the firm’s focus on training world models using gaming data.

The round and valuation are the clearest signals of the company’s runway: investors that joined range from venture firms to individual backers and researchers at major labs. The company also said it is building Nerve, a marketplace that connects gamers to data labeling and teleoperations work.

How does gaming data feed world models?

General Intuition argues large language models lack robust spatial and temporal understanding, and that world models trained on gaming data can fill that gap. The company is using gameplay and simulated environments to teach models how things move through space and time, aiming for agents that generalize beyond text.

The company points to an example where eight minutes of real-world data was enough to get a robot navigating an office cold, illustrating the payoff of compact, dynamics-rich datasets. General Intuition spun out of Medal TV, a gaming platform, and is channeling gameplay data and developer tooling into models intended for physical tasks and robotics.

What else is the company doing for the labor impact of its approach?

General Intuition is building Nerve, a marketplace that connects gamers to data labeling and teleoperations roles, which the company frames as a way to get ahead of AI-driven job displacement. The platform aims to create paid opportunities tied to the data and teleoperation work that underpins its world models.

The company also reportedly turned down an acquisition offer from OpenAI to remain independent, a decision the founders say preserves their mission and control as they scale.

Why it matters

If gameplay and simulated environments can teach models about physics, motion, and causality more efficiently than text alone, the industry’s path to physically capable agents could shift away from language-only training. That would reorient where training data value accumulates: from web text to interactive recordings and teleoperation traces.

The funding and investor mix — including Coatue, Eric Schmidt, and researchers from MIT and Google DeepMind — signal serious belief in that thesis and could accelerate the transfer of gaming-derived models into robotics, virtual agents, and other embodied systems. The company’s own moves, like Nerve, show an awareness of workforce impacts and an attempt to create paid work around dataset generation and labeling.

Ethics are already on the table: the founders discussed where the red lines lie when models trained this way could be used for defense applications, so scrutiny from policymakers and customers is likely to follow.

What to watch

Watch for the public rollout and uptake of Nerve as a concrete test of whether gamers will supply and monetize the kinds of data General Intuition needs. Also monitor partnerships or pilots that move world models from gameplay to physical robotics, and any disclosures about defense or dual-use applications, which the company flagged as an ethical concern.

Key data points to track: the company’s $320 million raise, its $2.3 billion valuation, investor composition including Coatue, Eric Schmidt, and researchers at MIT and Google DeepMind, and product milestones tied to Nerve and robot navigation tests.