General Intuition eyes $2B valuation for gaming-trained AI agents

Key Takeaways

• General Intuition is raising $300M at a $2B valuation, eight months after a $134M seed round
• The startup trains AI agents using 2 billion video game clips per year from Medal's 10 million monthly users
• Jeff Bezos, Eric Schmidt, Khosla Ventures, and General Catalyst are reportedly participating in the round

General Intuition, the New York startup training AI agents to understand physical space using video game footage, is in talks to raise $300 million at roughly a $2 billion valuation. That would mark a 15x jump from its $134 million seed round just eight months ago, according to sources who spoke to TechCrunch.

The round reportedly includes Jeff Bezos and Eric Schmidt, alongside existing backers Khosla Ventures and General Catalyst. If closed at these terms, General Intuition would become one of the fastest-rising AI startups of 2026, powered by a data asset most competitors cannot replicate.

What does General Intuition actually build?

The company spun out of Medal, a platform where gamers upload and share video clips. Medal processes about 2 billion videos per year from 10 million monthly active users. General Intuition feeds that footage into foundation models designed to teach AI agents spatial-temporal reasoning. The goal: machines that can perceive a 3D environment, anticipate what happens next, and act in real time.

Pim de Witte, who co-founded Medal, leads General Intuition. His co-founders include researchers Eloi Alonso, Adam Jelley, and Vincent Micheli, all of whom specialize in world modeling and simulation. The team's thesis is simple: first-person gameplay footage is ideal training data because it captures continuous human decision-making in dynamic environments. Language models learn from text. General Intuition's models learn from action.

Why gaming data matters for robotics and simulation

Traditional robotics labs collect training data from expensive physical robots or carefully staged simulations. General Intuition takes a different path. Gamers generate billions of hours of interactive footage for free, navigating mazes, dodging obstacles, coordinating with teammates, and reacting to unpredictable opponents.

That dataset reportedly caught OpenAI's attention. Sources say OpenAI previously tried to acquire Medal, and other major AI labs have also approached the company. The strategic interest makes sense: high-quality embodied data is scarce, and Medal's corpus of first-person gameplay is one of the largest in existence.

On HackerNews, discussion has focused on whether gaming footage can serve as a proxy for real-world robotic navigation. Skeptics argue that video games lack the physics fidelity robots need. Believers counter that the sheer volume and variety of human behavior in games provides a foundation that physical data collection cannot match at this scale.

How General Intuition differs from Runway and World Labs

The world model space has grown crowded. Runway, Decart, and World Labs have all shipped world models recently. Google's Genie 3 now integrates Google Maps data for real-world simulation. All of them see gaming and robotics training as near-term commercial applications.

General Intuition's business model diverges from the pack. Competitors often sell access to their world models. General Intuition uses world models internally to train agents, then sells the agents themselves. The company's pitch is that its unique dataset gives it a durable advantage in agent quality. If the model improves, every agent built on it improves. If the dataset keeps growing, the moat deepens.

What the company plans to do with $300 million

According to sources, General Intuition will use the funds primarily to scale compute capacity. The company aims to release a new product by late summer or early fall 2026. Details on what that product looks like remain thin, but the timeline suggests they are close to a commercial-ready agent offering.

The funding round, if finalized, would bring General Intuition's total raised to roughly $434 million in under a year. That pace of capital accumulation signals investor conviction, but also pressure. A $2 billion valuation demands either rapid revenue growth or continued dominance in a space where Google, OpenAI, and well-funded startups are all racing.

The investor lineup and what it signals

Jeff Bezos and Eric Schmidt have both made aggressive bets on AI infrastructure and robotics. Bezos recently led a $12 billion round for Prometheus, his artificial general engineer project. Schmidt has backed multiple AI startups through his foundation and personal investments. Their participation in General Intuition suggests they see embodied AI as a distinct, investable category separate from large language models.

Khosla Ventures and General Catalyst returning for a second round indicates the seed investors are satisfied with progress. Both firms have deep portfolios in enterprise AI and would not double down lightly on a speculative bet.

Frequently Asked Questions

What is General Intuition?

General Intuition is a New York-based AI startup that trains embodied AI agents using video game footage. It spun out of Medal, a gaming clip platform, in late 2025.

How does General Intuition train its AI agents?

The company uses 2 billion video game clips per year from Medal's 10 million monthly users. The first-person gameplay teaches AI spatial-temporal reasoning, allowing agents to perceive and navigate 3D environments.

Who is investing in General Intuition?

Sources report that Jeff Bezos, Eric Schmidt, Khosla Ventures, and General Catalyst are participating in the $300 million round.

How is General Intuition different from other world model startups?

While competitors like Runway and World Labs sell access to world models, General Intuition uses its models internally to train agents, then sells the agents as the final product.

When will General Intuition launch a product?

The company plans to release a new product by late summer or early fall 2026, according to sources familiar with the matter.

Source: TechCrunch / Rebecca Bellan