General Intuition raises $320M to train AI agents through video games

General Intuition, a London-based artificial intelligence startup, has secured $320 million in funding to develop AI agents trained on video game data, marking one of the largest European AI funding rounds this year. The company’s approach centres on using gameplay environments to create agents capable of generalising to real-world tasks.

The funding round, which values the company at $2.3 billion post-money, attracted participation from multiple institutional investors according to reports from TechCrunch and Reuters. General Intuition was co-founded by researchers from University College London, who developed the core methodology for extracting transferable intelligence from gaming scenarios.

The startup’s thesis addresses a persistent challenge in AI development: agents trained on specific datasets often fail when confronted with novel situations. By using video games—which contain complex decision-making scenarios, physics simulations, and adversarial interactions—General Intuition aims to create more robust models that can adapt to unfamiliar environments.

Video games provide several advantages as training grounds. They generate vast quantities of structured interaction data, present agents with clear objectives and feedback loops, and simulate consequences without real-world risks. The approach builds on academic research showing that game-trained models can develop reasoning capabilities that transfer beyond their original context.

However, the methodology faces scrutiny. Critics note that even sophisticated games remain simplified representations of reality, lacking the ambiguity, incomplete information, and social complexity that characterise real-world decision-making. The gap between simulated and actual environments has historically proven difficult to bridge in robotics and autonomous systems.

The business implications extend across multiple sectors. Companies developing autonomous vehicles, warehouse robotics, and industrial automation could benefit from more adaptable AI agents. Conversely, firms invested in traditional supervised learning approaches—which require extensive labelled datasets for specific tasks—may find their competitive position eroded if game-based training proves more efficient.

The gaming industry itself stands to gain. Studios could monetise their game engines and environments as training platforms, creating a new revenue stream beyond entertainment. This could accelerate development of more sophisticated game AI, as commercial incentives align with research objectives.

General Intuition’s funding also signals continued investor appetite for foundational AI capabilities despite market volatility. The company joins a cohort of well-capitalised startups pursuing novel training methodologies, including firms focused on synthetic data generation and reinforcement learning from human feedback.

The startup plans to deploy the capital towards expanding its computing infrastructure and recruiting researchers specialising in reinforcement learning and transfer learning. According to Startups Magazine, the company currently employs approximately 85 people, primarily based in London with a smaller presence in California.

Regulatory considerations may emerge as game-trained agents move into commercial deployment. Questions around liability, safety validation, and performance guarantees in critical applications remain unresolved. Law360 reported that the company has engaged with UK regulatory bodies regarding future deployment scenarios, though specific frameworks are still under development.

The competitive landscape includes both established AI laboratories and startups exploring alternative training paradigms. DeepMind, Google’s AI subsidiary, has published research on game-based training but has not commercialised a dedicated platform. Several Chinese firms are reportedly pursuing similar approaches, though with less public disclosure.

Market observers will watch for published benchmarks demonstrating real-world task performance from game-trained agents. The company has not yet released comparative data showing superiority over conventional training methods on standardised tests. Such evidence will prove critical in converting research promise into commercial adoption.

General Intuition’s substantial funding provides runway to pursue a technically ambitious agenda, but the approach must ultimately demonstrate practical advantages in cost, performance, or adaptability to justify its valuation and validate video games as a viable path to general-purpose AI agents.