General Intuition’s $6B Valuation Is a Bet on Gaming Data
A startup spun out of a gaming-clip app may soon be worth $6 billion before proving robots can work at scale. That is either brilliant—or the most expensive demo reel in venture capital.
A startup spun out of a gaming-clip app may soon be worth $6 billion before proving robots can work at scale. That is either brilliant—or the most expensive demo reel in venture capital.
The $6 billion question
General Intuition is reportedly in talks to raise new money at a $6 billion pre-money valuation, with Valor Equity Partners, Point72 Ventures and Seven Seven Six among the prospective new backers. Existing investors Khosla Ventures and General Catalyst are also said to be participating.
Let that sink in. The company raised $320 million at a $2.3 billion valuation only weeks earlier. Now it is chasing a valuation more than two-and-a-half times as high.
That is not normal startup progress. That is venture capital deciding that whoever owns the best “robot brain” could own a bloody big slice of the physical economy.
General Intuition is led by Pim de Witte and was spun out of Medal in October 2025. Medal is a platform where gamers upload and share clips. The clever bit is not the video itself. It is the underlying record of what players actually did: which buttons they pressed, when they pressed them, and what happened next.
That gives General Intuition something most AI companies would kill for: huge volumes of labelled action data. Not just pictures of a person climbing a ladder or moving through a virtual city, but the instruction trail behind it.
The company’s pitch is that this data can help train models to understand movement, cause and effect, space, time and action—then transfer some of that learning into simulations and eventually physical robots.
In plain English: ChatGPT learned to predict the next word. General Intuition wants machines to predict the next useful move.
That is why the money is arriving with a siren on top.
Why gaming data may be worth more than another chatbot
Most founders see data as a spreadsheet, a CRM export or a dashboard nobody opens. In AI, data is increasingly the factory.
Anyone can rent GPUs. Plenty of well-funded companies can hire former researchers from Google DeepMind, OpenAI or Meta. A decent product interface can be copied by lunch on Tuesday. But a proprietary dataset that captures millions of human decisions inside dynamic environments is a different beast.
Medal gave General Intuition access to hundreds of millions of hours of gameplay. More important, those clips came with action labels: the exact inputs that moved a player through a changing environment.
That is the company’s real asset. Not a quadruped robot wandering around an office. Not a flashy demo. Not another slide promising “embodied intelligence.” The asset is the connection between perception, decision and outcome.
A language model sees text and predicts text. A large action model needs to see a situation, infer what matters, choose an action and then deal with consequences. Walls do not care about your pitch deck. A robot that misjudges a wall, a ladder, a forklift or a human being creates a very expensive problem very quickly.
General Intuition’s argument is that games are a scalable training ground for this. Games have rules, obstacles, goals, changing scenes and constant feedback. Better still, they let a model make mistakes without smashing warehouse equipment or running over someone’s foot.
That is a serious advantage if it works.
The phrase worth underlining is if it works.
The gap between a Fortnite bot and a factory robot
Venture capital is behaving as though “physical AI” is the next obvious gold rush. The logic is not stupid. The world still runs on people driving, lifting, sorting, assembling, inspecting, cleaning and fixing things. Software has eaten much of the desk job. The physical world is the bigger meal.
But the real world is a bastard compared with a game.
Factories have poor lighting, broken sensors, impatient people, odd-shaped objects, rain, dust, stairs, insurance policies and customers who do not care that your model had an excellent benchmark score. A robot does not get to say it is having a bad data day when it is unloading a truck.
General Intuition has demonstrated that its model can move from gameplay and simulated environments toward real-world robotics. It has tested on a quadruped robot, as well as drones and other devices. The company says it needs only small amounts of physical-world data to fine-tune what it learned in games.
That is the entire investment case in one sentence: can cheap, abundant action data from games reduce the absurd cost and slowness of collecting real-world robot data?
If yes, $6 billion may look cheap in hindsight.
If no, this becomes another reminder that a model capable of looking clever in a controlled demo is not automatically capable of generating revenue in a dangerous, messy workplace.
The company itself is not pretending the job is done. It plans to spend heavily on compute and talent, with a CoreWeave partnership supporting its infrastructure. It has only a handful of customers across gaming, simulation and robotics. Its API was expected to become more broadly available by the end of the northern summer.
So this is not a mature business being valued on proven earnings. It is a data advantage being priced as a future platform.
That distinction matters, especially for founders who mistake a hot valuation for a finished company.
The overlooked angle: this is a supply-chain bet, not just an AI bet
Everyone is staring at the model. I am staring at the supply chain.
A useful physical-AI company needs far more than an impressive foundation model. It needs compute. It needs hardware partners. It needs customers willing to let new technology operate around expensive assets and real staff. It needs deployment capability, safety processes, support, and a way to collect valuable new data without creating legal or operational chaos.
That is why General Intuition’s stated ambition to be the backbone for other builders is more interesting than trying to build every robot itself.
Building the full stack is seductive. You get to show off the robot, control the demo and pretend vertical integration is a personality trait. It is also a fantastic way to burn money across hardware, software, manufacturing and sales before you have nailed any of them.
General Intuition’s better strategic choice may be to sell the intelligence layer to companies building robots, simulations and specialised machines. That puts it closer to the picks-and-shovels end of the market.
But there is a catch. Platform businesses only win when developers and customers actually build on them. The company will need to prove that its model performs across enough settings to justify becoming infrastructure rather than a very well-funded research lab.
That is where the next round of evidence needs to come from: repeatable customer outcomes, not more investor logos.
The contrarian view: the valuation is not the scary part
People will look at the jump from $2.3 billion to $6 billion and yell “bubble.” Maybe. But valuation is not the main risk here.
The bigger risk is that every company in this race learns the same lesson at once: models need proprietary action data, and the easy sources are already owned.
General Intuition has Medal. Other players have industrial fleets, driving data, robot teleoperation networks, simulation environments or deep customer relationships. The market may not be won by whoever has the cleverest model this quarter. It may be won by whoever creates the best data flywheel over the next five years.
That means the founders who survive will be the ones who design their business so every customer deployment makes the product harder to compete with.
That is a much more durable strategy than raising money because the market currently enjoys the phrase “robotics AI.”
I also like that de Witte has drawn a line against lethal autonomy. Whether you agree with every boundary or not, it is a reminder that founders should decide where their technology goes before a giant customer makes the decision for them. Principles are cheap on a website. They get expensive when there is a nine-figure contract on the table.
What this means for you
If you are a founder, stop asking whether you can add AI to your business. That question is already tired.
Ask these four better questions tomorrow morning:
1. What proprietary behaviour data do we create? Not generic customer records. What do users repeatedly do inside your product that reveals decisions, intent and outcomes?
2. Does each use of our product improve the next use? If the answer is no, you may have a feature, not a compounding advantage.
3. Where does simulation save us money before reality punishes us? You do not need robots to use this principle. Test pricing, workflows, onboarding and sales processes cheaply before rolling them into the real operation.
4. What would make a customer trust this with consequential work? Accuracy is only one part. Reliability, integration, accountability and support matter just as much when a product touches money, safety or a real operational bottleneck.
If you are an investor, do not get hypnotised by the valuation. Find the data source, understand whether it is exclusive, and ask how it gets stronger after the first cheque is spent.
And if you are an operator, watch this space without getting carried away. The first useful wave of physical AI will probably not look like a humanoid robot replacing an entire team. It will look like a narrow task done more safely, more cheaply or more consistently than before.
That is how real businesses are built: one painful job at a time.
General Intuition may become a giant. Or it may discover that gaming data is brilliant right up until the moment a robot meets a warehouse floor. Either way, the lesson is already clear: in the next phase of AI, the scarce asset will not be clever words. It will be proof that a machine can act without stuffing it up.