General Intuition has closed a $320 million Series B round led by Khosla Ventures, pushing the company's valuation to $2.3 billion, according to GamesBeat and TechBuzz.ai, both reporting on 25 June 2026.

The capital is directed at a specific thesis: that millions of hours of gameplay data constitute a high-quality training substrate for frontier AI agents. Where most large-scale agent training pipelines lean on web text, synthetic data, or robotic teleoperation, General Intuition is betting that video games offer something those sources do not — richly structured, causally dense environments in which an agent must plan, adapt, and act under tight feedback loops. The argument, as TechCrunch frames it, is that this translates to more capable real-world agents.

The intuition behind the intuition, so to speak, is not new to the research community. DeepMind's work on AlphaGo and AlphaStar, and OpenAI's early Dota 2 experiments, both established that game environments can produce emergent strategic reasoning that transfers — at least partially — beyond the game itself. What General Intuition appears to be doing is operationalising that principle at commercial scale, with a corpus of gameplay hours large enough to train frontier-class models rather than narrow task-specific ones.

That is a non-trivial architectural and data-engineering challenge. Gameplay data is high-dimensional and heterogeneous: frame-level visual observations, controller inputs, in-game state representations, and outcome signals all need to be aligned and tokenised in ways that generalise across titles, genres, and player skill levels. Getting that pipeline right is arguably as hard as procuring the data in the first place — and a $320 million raise suggests the company believes it has, or can build, the infrastructure to do it.

Worth noting on the funding structure: Khosla Ventures leading a Series B at this valuation places General Intuition firmly in the tier of AI infrastructure bets the firm has made alongside companies like Mistral and Perplexity. A $2.3 billion post-money on what is still a pre-revenue or early-revenue company is consistent with how frontier AI has been priced across the board since 2024, but it concentrates substantial execution risk on a training methodology that has not yet been independently validated at production scale.

The broader question the company will face is transfer. Game environments are internally consistent and fully observable in ways the physical and digital world is not. An agent that masters the resource management of a real-time strategy title or the spatial navigation of a first-person game has operated in a simulator — one far richer than most synthetic benchmarks, but a simulator nonetheless. The degree to which game-trained representations encode genuine world models rather than domain-adapted heuristics is an open empirical question, and one the research community will scrutinise closely once General Intuition begins publishing results.

That scrutiny aside, the funding confirms that well-capitalised investors are willing to back heterodox training data strategies at scale. The dominant paradigm — ever-larger crawls of internet text — faces well-documented saturation concerns, and the field is actively hunting for high-signal alternatives. Gameplay data, with its built-in reward structure and action-observation sequences, fits the template of what reinforcement learning from human feedback researchers have been trying to synthesise artificially. General Intuition's argument is that the real thing, captured from human players across decades of titles, is better.

Whether the $2.3 billion valuation is eventually justified will depend on whether agents trained on gameplay data outperform those trained by other methods on tasks that actually matter to enterprise and consumer buyers. The next data point will likely come from model evaluations or a product launch — neither of which has been publicly announced as of 25 June 2026.