World model funding: General Intuition at $6.2B, Zenithon $10M

Two bets on world models, at very different scales

Two funding announcements this week show how much investors want AI that can model, predict and act in physical environments. They also show how differently that bet can be sized.

General Intuition, an AI research lab that trains foundation models on gameplay footage, raised $220 million at a reported $6.2 billion valuation. 45 The backers included Valor Equity Partners, Atreides, Seven Seven Six (776), Point72, Khosla Ventures and General Catalyst. 34 At the other end of the range, London-based Zenithon AI disclosed $10 million to build world models for fusion reactors, rockets and semiconductor manufacturing. 125

The two deals sat within a broader daily batch. Across ten selected companies, newly disclosed financing came to roughly $280.6 million in dollar-denominated capital plus €7.6 million in euro funding. General Intuition accounted for the large majority of that figure. 5

General Intuition: the data is the moat

General Intuition says it is building "an entirely new class of foundation models." These are trained on billions of action-labeled videos and designed to act in real time in environments they haven't seen before. 3 The key detail is the labeling. Each clip pairs the on-screen footage with the keyboard or controller inputs that produced it. 4 That gives the models a direct link between what an agent sees and what it does. The company says it designs these models to carry over from games to simulation, and then to physical systems such as robots, drones and vehicles. 4

The data comes from Medal, General Intuition's sister company and a clip-sharing platform where gamers upload highlights. Medal is reportedly on pace for three billion uploads a year, a supply that competitors would struggle to match. 3 One funding roundup called this the clearest signal of the current cycle. In its framing, investors are paying for proprietary data that is hard to reproduce rather than for ownership of a generic foundation model. 5

There is some inconsistency in the reported numbers. One market-data profile lists the round as a Series B and puts total funding at about $673.7 million across three rounds. 4 Another write-up headlines the $6.2 billion figure but cites a $4.01 billion valuation in its body text. 3 The $6.2 billion number is the one most consistently reported. Even so, readers should treat the exact valuation as somewhat unsettled until the company confirms the terms. General Intuition has also opened a waitlist for partners who want early access. 3 That suggests commercial deployment is still at an early stage.

Zenithon: replacing months of simulation with seconds

Zenithon's pitch is narrower and more technical. Simulating plasma behavior in a fusion reactor can take months, and by the time results arrive, engineers have often already changed the design. 1 Co-founder Alex Higginbottom saw that gap while working on a machine-learning PhD. He teamed up with Abetharan Antony, a plasma physicist who spent years working through tokamak calculations by hand. 1 The pair founded the company in 2025 after working together in fusion research. 2

The company is building neural-operator world models that aim to cut simulation times from months to under a second. 12 Its stated goal is to test a million engineering designs in the time a single conventional simulation used to take. 1 Backed VC, Lunar Ventures, Seraphim Space, MMC Ventures and SOSV led the round, with angel investors also taking part. 12 One outlet describes it as a seed round. 1 The investor mix fits the target markets. Seraphim's space focus, for example, lines up with Zenithon's rocketry ambitions.

Reading the signal

It is tempting to treat both companies as part of one "world model" wave. Their strategies actually sit on opposite sides of a key divide.

General Intuition is betting on scale. It wants very large amounts of behavioral data to produce general agents that can then specialize. Zenithon is betting on depth. It is building models for physics regimes where errors are costly and expert knowledge is scarce. That second category is becoming crowded. Zenithon will need to show that its models hold up on real reactor and engine designs, not just on benchmarks.

What connects the two deals is the idea of a defensible edge. For General Intuition, that edge is Medal's data pipeline. For Zenithon, it is founders with deep fusion experience aimed at a clear industrial bottleneck. The broader funding picture supports the same conclusion. 5 Capital is flowing to AI companies that control something rivals cannot easily copy, and less to those offering general-purpose capability alone.

The next question is transfer. Both companies are promising that what works in simulation will hold in the physical world: games to robots in one case, neural approximations to real plasma in the other. Their valuations will depend on whether they can deliver that.