General Intuition raises $220 million at a $6.2 billion valuation, betting gameplay clips can teach robots to act
General Intuition, which trains AI models on action-labelled gameplay clips from the Medal platform, said on 29 September 2026 that it had raised $220 million at a $6.2 billion valuation from investors including Khosla Ventures and General Catalyst. The figure is 2.7 times the $2.3 billion it disclosed with its Series A 96 days earlier, and the company has opened a waitlist for partners in robotics, simulation and games.
Ninety-six days after announcing a $2.3 billion valuation, General Intuition has nearly tripled it. The New York-based AI lab said on 29 September 2026 that it had raised $220 million at a $6.2 billion valuation, on the argument that clips of people playing video games can teach machines, robots included, how to act in places they have never seen.
The round matters to robotics because General Intuition builds no robots. It owns no arm and no humanoid. What it says it owns is the largest supply of action-labelled video in the industry, and the open question is whether that data can stand in for the slow and costly teleoperation on which robot foundation models now depend.
Six investors, no lead and no stated basis for the price
The company named Valor Equity Partners, Atreides, 776, Point72, Khosla Ventures and General Catalyst as investors in a thread on its official X account. It did not name a lead or a series label, and it did not say whether $6.2 billion is measured before or after the new money. The proceeds will pay for a larger team in New York and Europe, the company said.
The raise follows a $320 million Series A at a $2.3 billion valuation, which General Intuition announced on 25 June 2026 with Khosla Ventures leading and General Catalyst, Eric Schmidt and Jeff Bezos taking part. Its website now says it has raised more than $650 million in the past year. The two rounds it itemised account for $540 million, so more than $110 million came in earlier.
Action models and world models, sold through an API
General Intuition calls itself a lab for acting in space and time. It builds action models, which choose the next move from pixels, and world models, which predict what happens after a move. Its flagship world model, MIRA, was trained on 10,000 hours of Rocket League play and runs in real time at 20 frames per second, according to the company website, which credits the lab Kyutai and Epic Games as collaborators. The company says select partners in gaming, simulation and robotics already use a commercial API, and with the new round it opened a waitlist for early access.
Why a game controller looks like a teleoperation rig
The technical bet is simple to state. A player steering a car in a game and an operator driving a robot do the same thing: they watch a screen and move a controller. Every clip on Medal, the gameplay-sharing platform that General Intuition's website describes as its foundation, therefore carries a record of which inputs were pressed as the picture changed. The company compares this with how Tesla trained its driving software, across many more actions and settings. It says Medal is on track for 3 billion uploaded clips a year, that leading robotics and world models are trained on less than 1% of the action data it can reach, and, in June, that the clips came from 17 million monthly active users.
The busiest week for world models so far
The raise lands in a crowded week. AMD agreed on 28 September to buy World Labs for about $8.2 billion in stock, and Runway released Praxis-1, an open-weight world action model for robot control, on 30 September. NVIDIA opened its Cosmos 3 Edge world model for on-robot planning in July. Skild AI, which raised $1.4 billion at a $14 billion valuation in January, and Physical Intelligence train robot foundation models on robot data. General Intuition differs in where its data comes from. Robot-first companies collect demonstrations on the hardware they target; General Intuition collects them from players and still has to show that the skills carry over to real machines.
ROBOTNESS analysis
General Intuition is being priced as a data company more than a model company: investors are paying for privileged access to action-labelled video at a scale no robot fleet can match, and the valuation holds only if that data measurably shortens the road to working robots.
The pace of the step-up makes the point. On the valuations the company stated, its price rose 2.70 times in 96 days, while Skild AI's rose 9.3 times over the 554 days between its Series A and Series C. General Intuition has disclosed no revenue, no named robotics customer and no result on physical hardware. What changed between June and September is that it began offering models to outside partners, and that rivals were bought or priced in the billions.
- General Intuition
- Action and world models from gameplay
- Unnamed
- 220
- 6.2
- 2026-09-29
- Skild AI
- Robot foundation model
- Series C
- 1400
- 14
- 2026-01-14
- Figure AI
- Humanoids and Helix model
- Series C
- 1000
- 39
- 2025-09-16
- SiMa.ai
- Edge AI chips for physical AI
- Series C
- 150
- 1.45
- 2026-09-28
- Humanoid
- Humanoid robots
- Series A
- 152
- 1.35
- 2026-07-21
- Walden Robotics
- General-purpose robots
- Seed
- 300
- 1.1
- 2026-07-15
Valuations as disclosed by each company. General Intuition did not state whether its figure is pre- or post-money; the others are post-money as disclosed. World Labs is excluded because AMD's roughly $8.2B is a stock purchase price, not a funding valuation. Figures disclosed, not estimated.
As of Oct 1, 2026
The strongest counter-argument is that a controller is not a gripper. Games offer clean state, consistent physics and no contact forces, while a robot hand slipping on a wet bottle offers none of them. The claim that the models act in unseen environments, including the real world, is so far the company's own, with no reproducible test behind it.
- 2026-06-25
- Series A
- 320
- 2.3
- Khosla Ventures
- 2026-09-29
- Unnamed
- 220
- 6.2
- No data
Step-up = 6.2 / 2.3 = 2.70x over 96 days. Skild AI comparison: 14 / 1.5 = 9.3x over 554 days (Series A 2024-07-09 to Series C 2026-01-14). Company says it raised more than $650M in the past year; itemised rounds total $540M. Figures disclosed, not estimated.
As of Oct 1, 2026
In the bull case, robotics partners on the API report faster training with General Intuition's models as a pre-training layer, a named robot maker adopts them by mid-2027, and Medal's flow of clips becomes data that robot-first rivals must license. In the bear case, transfer from game physics to contact-rich manipulation stays weak, robot-first labs keep their lead because their data comes from the target hardware, and video-model companies such as Runway close much of the data gap.
Three dated signals will confirm or break the thesis.
- By 31 March 2027, whether General Intuition names a robotics customer or publishes a result on physical hardware.
- By 30 June 2027, whether the September waitlist turns into general availability with published pricing.
- Through the first half of 2027, whether robot-data rivals such as Skild AI or Physical Intelligence raise at valuations that widen or close the gap.
- $220 million
- 29 September 2026
- Valor Equity Partners, Atreides, 776, Point72, Khosla Ventures, General Catalyst
- $6.2 billion (company did not state pre- or post-money)
- Medal gameplay clips; company says Medal is on track for 3 billion uploads a year
- Not named
- $320 million Series A at $2.3 billion, 25 June 2026, led by Khosla Ventures
- 2.70x in 96 days (ROBOTNESS calculation)
- More than $650 million (company website)
- MIRA, trained on 10,000 hours of Rocket League play, 20 fps in real time
Why it matters
Robot foundation models are limited less by architecture than by the supply of demonstrations, which today come mostly from teleoperation on real hardware. General Intuition is the first well-funded company to argue that consumer gameplay, recorded with exact controller inputs, can fill that gap at internet scale. If even a fraction of that data transfers to physical control, the economics of robot pre-training change, and the scarce asset moves from robot fleets to data platforms.
The round also sets a reference price. At $6.2 billion, a company with no disclosed revenue and no robot hardware is valued above every private robotics company with a disclosed valuation in the ROBOTNESS database except Figure AI and Skild AI. That pricing will be used by investors in other world-model and data companies over the next two quarters.
Rival analysis
Robot-first model builders such as Skild AI and Physical Intelligence own data from the hardware they target, which is their advantage on contact-rich manipulation. Video-model companies such as Runway, whose Praxis-1 world action model came out on 30 September, and NVIDIA with Cosmos, start from passive video without action labels and must infer actions. General Intuition sits between them: its video carries real action labels, but from virtual bodies.
World Labs, which AMD agreed to buy for about $8.2 billion in stock, builds 3D world generation rather than action models, so it is a neighbour more than a direct rival. The most direct competition may come from any platform that owns large logs of human control input, from game publishers to driving-simulator makers, if they decide to train or license.
Valuation context
The company's stated valuation rose from $2.3 billion to $6.2 billion between 25 June and 29 September 2026, a 2.70x step-up in 96 days. Skild AI's disclosed post-money rose 9.3x from $1.5 billion to $14 billion over 554 days. Measured per dollar raised in 2026, General Intuition's $6.2 billion sits on $540 million of itemised capital, about 11.5 times, against Skild AI's $14 billion on a $1.4 billion Series C, 10 times.
The basis of General Intuition's figure is not stated, and the company has not disclosed revenue. We therefore treat these ratios as indicators of investor appetite rather than of fundamentals, and we do not compare them with public-market multiples.
Supply-chain implications
General Intuition's main input is data it already controls through Medal, plus GPU capacity for training and serving. It competes for the same accelerator supply as every other frontier lab. The company has no hardware bill of materials and no exposure to actuators, reducers or batteries.
The dependency to watch is Medal itself. The data advantage rests on a consumer platform whose growth, user terms and privacy rules sit outside the robotics market, in the US and the EU in particular.
Signals to watch
First, a named robotics customer or a published physical-hardware result by 31 March 2027. Second, general availability and pricing for the API by 30 June 2027. Third, the next rounds of robot-data rivals, which will show whether investors keep paying a premium for gameplay data over robot data. A fourth, softer signal is hiring: new robotics-specific roles in New York or Europe would show where the company expects revenue.
Analyst view
Thesis: General Intuition is valued for privileged data access, and the valuation depends on proof that gameplay-trained policies transfer to robots. Confidence: low to medium. The data scale and the investor list are concrete, but there is no public physical benchmark, no revenue and no named robot customer.
We would raise confidence on a published sim-to-real result with a robot partner, and lower it if the first robotics customers use the models only for simulation or entertainment rather than control.
Questions you should be asking
What share of current API partners are robotics companies, and are any running General Intuition models on physical hardware? Is the $6.2 billion pre- or post-money? How are Medal users informed about and compensated for the use of their clips in commercial models? What licensing terms will robot makers get, exclusive or open? How does the company plan to handle contact forces and proprioception that game data does not contain?