이미지: METAL LAB 생성
Summary
- General Intuition is in talks for a new funding round at a $6 billion pre-money valuation.
- That's nearly triple the $2.3 billion valuation it had just weeks ago, with Valor Equity Partners and others joining as new investors.
- Spun out of the gameplay clip platform Medal, the company uses gameplay data to train robot movement.
- 신규 밸류에이션
- 60억달러 프리머니, 협상 진행 중
- 신규 투자자
- 밸러 이쿼티 파트너스, 포인트72 벤처스, 세븐세븐식스
- 재참여 기존 투자자
- 코슬라벤처스, 제너럴 카탈리스트
- 직전 라운드
- 3억2000만달러 유치, 23억달러 밸류에이션(몇 주 전)
- CEO
- 핌 드 위테, 지난해 10월 메달(Medal)에서 분사해 창업
- 초기 데이터셋
- 게임플레이 영상 수억 시간 + 버튼 조작 기록 '액션 라벨'
- 컴퓨팅 파트너
- 코어위브(CoreWeave)
Valuation Triples in a Matter of Weeks
New York-based General Intuition is in talks for a new funding round. The pre-money valuation on the table is $6 billion — a striking jump considering the company raised $320 million at a $2.3 billion valuation just weeks earlier. That's nearly a threefold increase in a short span. New investors joining this round include Valor Equity Partners, Point72 Ventures, and Seven Seven Six, while existing backers Khosla Ventures and General Catalyst are putting more money in as well.
From Gameplay-Clip Company to Physical AI Company
General Intuition is a startup building foundation models that teach general-purpose AI agents how to move through space and time. CEO Pim de Witte spun the company off last October from Medal, the gameplay-clip sharing platform he had been running. Hundreds of millions of hours of gameplay footage on Medal, along with "action labels" recording which buttons players pressed and when, served as the initial training data. The idea is to translate in-game control inputs into movement for physical robots with actual bodies.
This TechCrunch report corroborates an earlier Wall Street Journal report, and the company is now considered one of the hottest startups in physical AI. De Witte said in a post on the 21st that he had shown the Wall Street Journal, for the first time, how the company's large action model actually works.
The Money Is Headed Toward Robotic Embodiment
The round hasn't closed yet, and sources familiar with the deal say it's oversubscribed given the level of investor interest. The company plans to use the funding to improve its general-purpose model with a focus on robotic embodiment. That means ramping up spending on compute infrastructure and expanding hiring — on the compute side, it already has a partnership in place with CoreWeave.
Khosla Ventures' Vinod Khosla said action-label data is central to what he called "the emergence of intuition" — the ability of a model to generalize to tasks it wasn't explicitly trained on.
Valor Equity Partners is known for its investment in SpaceX, and if this deal closes, it would mark the fund's first AI lab investment since SpaceX. TechCrunch said it had reached out to Valor for confirmation.
Editor's Take
What stands out about this round is the sheer speed of the valuation climb. Startups doubling or tripling their valuation within months hasn't been unusual since the generative AI boom took off, but a threefold jump in a matter of weeks is rare even by those standards — especially in physical AI, a field centered on robot bodies rather than software. While language-model companies like OpenAI and Anthropic built their valuations on text and code, General Intuition is hitting a similar pace with an entirely different kind of data: recorded game-controller inputs.
It's also worth noting that this would be Valor Equity Partners' first AI lab bet since SpaceX. A fund that has built its reputation on hardware and physical systems opening its wallet for a gameplay-data-driven robotics model is a signal that the investment world's yardstick for "embodied AI" is shifting.
There's something here worth watching for Korea's gaming and robotics industries too. The idea of repurposing game logs for robot training means that any game company sitting on hundreds of millions of hours of control-input records already holds a physical-AI dataset — no separate simulation build required. That said, the physical constraints of a real robot joint are different from virtual controls in a game, so how well this data actually transfers to real robot arms or bipedal robots remains a separate question that still needs to be tested.
The round remains oversubscribed and unclosed. Once it does close in the coming weeks, we'll find out both whether Valor actually participates and what kind of robotics partnerships this funding ends up fueling.
Comments