General Intuition raises $320M at $2.3B to train agents on gameplay action data

General Intuition announced a $320 million Series A at a $2.3 billion valuation on June 25, 2026 (round closed in January). Total funding now stands at $454 million. The round was led by Khosla Ventures with participation from General Catalyst, Hillspire (Eric Schmidt), Jeff Bezos, Nico Rosberg, and researchers from Google DeepMind and MIT.

The startup spins out from Medal, a gaming clip platform with 17 million monthly active users uploading roughly 2 billion gameplay clips annually. Crucially, each clip embeds ground-truth action labels: the exact buttons pressed and timing that produced each frame. This differentiates General Intuition's approach—it trains action foundation models on real human intent under pressure, not just video inference.

General Intuition has demonstrated a quadrupedal robot navigating real office space after only 8 minutes of real-world fine-tuning, using the same embodied agent model trained on gameplay. The company is targeting a product API launch by end of summer 2026 and is already in discussions for a Series B. It also launched Nerve, a jobs marketplace paying gamers for gameplay footage and robot teleoperation data.

For architects: this signals the consolidation of embodied AI around synthetic-but-grounded data. Unlike pure simulation (brittle transfer) or expensive real-world robotics data collection, gameplay with action labels splits the difference—millions of hours of human control sequences annotating physical causality at scale. The theta on transfer risk is real, but investors are betting that action-level supervision beats pixel-only approaches for training agents that must perceive, predict, and act in physical space.