🌏 中文版
Deal Terms
| Item | Value |
|---|---|
| Company | General Intuition (New York, US) |
| Round | A growth round following its Series A (no official letter name reported) |
| Amount | $220M |
| Lead | Valor Equity Partners, Atreides Management (co-lead) |
| Participants | Seven Seven Six (776), Point72, Khosla Ventures, General Catalyst |
| Valuation | $6.2B |
| Total raised | $673.7M |
| Founded | 2025 (spun out of gaming video platform Medal) |
| Headcount | Undisclosed |
What the Company Does
General Intuition builds foundation models for the physical world — rather than training primarily on text or static web data like most AI labs, it focuses on action: robotics, physical environments, and spatial reasoning tasks that require responding to unfamiliar situations in real time.
Its core data advantage comes from its parent, Medal, a gaming-clip platform where players upload highlight footage — on track for roughly 3 billion uploads a year. The company argues that leading robotics and world models today train on less than 1% of the action data it can draw from Medal, and that gaming environments are uniquely useful because players constantly encounter unpredictable situations and react in real time; the company says its simulated environments see more "accidents" in a single day than occur across the entire United States. That data trains what the company calls General Agents and World Models — models meant to respond sensibly in environments they've never seen before. It released a research demo model called MIRA in June and is now testing a commercial version with a limited set of customers while opening a waitlist.
What This Round Signals
What It Means for the Agent Ecosystem
This round pushes the thesis that "agents don't need more text, they need more action data" all the way into a valuation. Most agent companies compete on how well they use existing large language models; General Intuition is betting upstream — if physical-world action data is itself the scarce resource, whoever holds the largest, most diverse action dataset has a shot at becoming the foundation-model supplier sitting furthest upstream in the robotics and embodied-AI supply chain.
What Investors Are Betting On
Lead investors Valor Equity Partners and Atreides Management used the same pairing just weeks earlier to price another physical-world AI startup, Flow Engineering (an agentic hardware-design platform), at $750M. Both bets point to the same thesis: the two funds are systematically treating "applying agentic and foundation-model techniques to physical-world problems" as a single theme to back, rather than making isolated, unrelated calls — though that could also mean pricing in this corner of the market is being set by a small group of investors cross-referencing each other's term sheets rather than a broad market consensus.
Numbers Worth Watching
- A valuation that has climbed to $6.2B on $673.7M in total funding since its Series A is among the largest single bets in the "world model" category, which still has no clearly shipped commercial product
- The company's claimed "roughly 3 billion video uploads a year" would be a scale no public robotics training dataset currently matches — but that figure comes only from the company's own disclosure and hasn't been independently verified
- This is the company's first push toward commercialization since releasing its research demo MIRA in June — scaling to real paying customers is still a distance away
Watchlist Status
General Intuition is not currently on the watchlist, and none of the existing categories (foundation models, agent toolchain, vertical agents) cleanly fit an emerging sub-category like "physical-world foundation models / world models." Recommend waiting to see whether enough comparable companies emerge (such as Flow Engineering's physical-world agentic approach) before deciding whether the watchlist needs a new "Physical AI / World Model" section, rather than forcing this into an existing one.
Today's Takeaway
Most AI funding stories are about squeezing more out of existing data with more compute. General Intuition flips that: when everyone is competing for the same pool of web text and video, what's actually scarce is data showing how humans improvise when something goes wrong — and gaming, usually dismissed as a side effect of entertainment, turns out to be the cheapest, largest-scale channel for hoarding exactly that.
References
Loading...