Valor, Point72 back General Intuition at $6B valuation as AI startup pushes into robotics | TechCrunch

General Intuition, a New York-based startup that trained its first AI models on video game footage, is closing a round at a $6 billion pre-money valuation backed by Valor Equity Partners, Point72 Ventures, and Seven Seven Six — a 2.6x jump from the $2.3 billion valuation it held just weeks earlier after a $320 million raise.

What's Converging

The market for physical AI — models designed to control robots and other embodied systems that act in the real world — has been absorbing capital faster than most infrastructure segments, driven by a problem that every serious player in the space shares: there isn't enough high-quality training data to teach robots how to move reliably across varied, unstructured environments. Gathering that data through physical robot trials is expensive and slow. The pressure to find a synthetic or proxy data source has pushed several labs toward simulation, video datasets, and now gaming footage as plausible stand-ins for real-world behavioral supervision.

At the same time, the investor class has shifted. Funds that historically stayed in software-only AI bets are now moving into embodied-AI startups with growing urgency. Valor Equity Partners, the firm most associated with SpaceX, is participating in General Intuition's current round — multiple reports describe this as Valor's first investment in an AI lab. Point72 Ventures, the venture arm of the multi-billion-dollar hedge fund, is joining alongside Seven Seven Six, with existing investors Khosla Ventures and General Catalyst also participating. That combination of aerospace-adjacent capital and institutional hedge-fund money entering the same robotics round says something about how the sector is being perceived at the highest levels of capital allocation.

A third signal worth noting: the emergence of "large action models" as a named category distinct from large language models. Where language models predict the next token in a text sequence, the LAM framing describes models that predict the next discrete action in a behavioral sequence — a distinction with real engineering implications for robot control, where the output space is physical motion rather than words. General Intuition is among the first startups to explicitly brand its architecture around this framing, and investors appear to be pricing in the possibility that LAMs develop their own competitive moat, separate from the text-based AI landscape.

The Specific Development

General Intuition CEO Pim de Witte spun the company out of Medal, his video game clip-sharing platform, in October 2025. Medal had accumulated hundreds of millions of hours of gameplay footage — not unusual for a gaming platform of its scale. The unusual part is that Medal also stored what de Witte calls "action labels": timestamped records of exactly which buttons a player pressed and when, paired to the on-screen context at that moment. Most gaming platforms log the video; Medal logged the control inputs alongside it.

De Witte's decision to use that action-labeled dataset as the foundation for a physical AI model is the core intellectual bet behind General Intuition. Vinod Khosla, whose firm is an existing investor now participating again, has framed this publicly as a path to what he calls the "emergence of intuition" — the ability for a model to generalize across tasks it was never explicitly trained on, using the structured behavioral supervision that action labels provide. Whether that generalization holds for real-world robotic tasks remains to be demonstrated at scale, but the framing has clearly moved institutional capital. The Endroid News Network analysis notes that, unlike raw video, action labels give models ground-truth behavioral supervision rather than passive observation — a meaningful distinction when the goal is teaching a system to act, not just perceive.

The round's structure is notable in itself. General Intuition closed that $320 million at $2.3 billion, and within weeks the pre-money valuation nearly tripled. Multiple reports confirm the round is oversubscribed, meaning investor demand exceeded available allocation. The company intends to apply the capital toward improving its foundation model with a specific focus on robotic embodiment — translating the gaming-trained action models into control systems for physical robots, with a CoreWeave compute partnership supporting the infrastructure side.

Our read is that the oversubscribed condition, combined with Valor's participation specifically, is the most meaningful signal to parse here. Valor's SpaceX track record suggests the firm is comfortable backing hardware-intensive bets that pure software VCs typically pass on early. Entering an AI lab at a $6 billion pre-money valuation — before General Intuition has demonstrated commercial robotic deployments at scale — is a significantly larger leap than backing a launch company with a physical rocket program already underway. It suggests Valor is reading the physical AI deployment timeline as materially closer than the consensus.

What's Likely Next

The most immediate question is whether the round closes at the reported $6 billion terms and who fills the remaining allocation. With the round oversubscribed, General Intuition has leverage to be selective, and the final investor list could reveal where the company expects to find its first commercial distribution — defense, manufacturing, and consumer humanoid robotics all represent different bets about which sector adopts general-purpose robotic control first. The 30-day window will likely bring either a formal announcement confirming the terms or a revised figure reflecting final negotiations.

The 90-day question is more interesting: can General Intuition show that action-labeled gaming data actually transfers to physical robotic control in a way that is measurable and reproducible? The Wall Street Journal ran what was described as a first look at how GI's large action models work in August 2026, creating public expectations the company now needs to meet with hardware demonstrations, not just architecture descriptions. If the company can show a robot completing a generalized manipulation task using a model that was never explicitly trained on that exact scenario, the $6 billion figure will look conservative in retrospect. If the gaming-to-robotics transfer proves harder to achieve than the investors are pricing in, the valuation velocity will face scrutiny that oversubscribed rounds typically postpone rather than prevent.

Sources