🎧 Prefer to listen?
I covered General Intuition’s initial $134M raise last month — the startup that trains AI agents using billions of video game clips from Medal’s 10 million monthly users. At the time, the pitch was interesting but unproven. Now they’ve raised another $320 million at a $2.3 billion valuation, and the story has gotten a lot more concrete. They’ve got a working robot, a jobs platform for gamers, and Vinod Khosla calling it a “generational bet.” Here’s what actually changed.
The $320M round: who’s betting on gameplay as training data
General Intuition’s latest round was led by Khosla Ventures, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers from Google DeepMind and MIT. That’s a lot of smart money for a startup that’s less than a year old.
The total disclosed funding now sits at $454 million. The company says the vast majority of the new capital will go toward scaling compute capacity — they have a deal with CoreWeave and plan to focus on pre-training the next version of their model. A slice is earmarked for making their API more broadly available by the end of summer.
What caught my attention isn’t the dollar amount — it’s who’s writing the checks. When the same people who backed OpenAI and Anthropic are putting money into a company that learns from Fortnite, it signals something about where the AI industry thinks the next breakthrough lives. And it’s not in bigger language models.
From video game clips to walking robots
Here’s the part that made me rethink what this company actually is. During a visit to their New York R&D floor, TechCrunch reporters watched an AI agent play Fortnite for 100 hours straight — then saw the same model power a quadrupedal robot walking around the office. The robot navigated using a single camera, bumped into chairs like a toddler, and fine-tuned itself to a new environment in just eight minutes of real-world data.
The key insight: the same brain that learns to navigate a virtual world can transfer that understanding to the physical one. This is what AI agents becoming employees looks like in practice — not chatbots answering emails, but models that understand space, time, and cause-and-effect well enough to walk a robot through an office.
Most competitors try to infer actions from video alone. General Intuition’s edge is the action labels embedded in Medal’s gameplay clips — records of exactly what buttons a player pressed and when. CEO Pim de Witte argues this helps the model distinguish “self” from “environment” in a way that gives it a richer understanding of causality. It’s the difference between watching someone drive and actually feeling the steering wheel.
Nerve: a jobs platform for the gaming generation
This is the part nobody’s talking about, and it might be the most important thing General Intuition is building. They recently launched Nerve, a jobs marketplace that lets gamers earn money using their existing hardware. Users start with data labeling tasks and can eventually move toward robot teleoperation and other higher-skilled work.
The logic is straightforward: Medal’s user base is precisely the generation most exposed to AI-driven displacement. De Witte wants them to have a stake in what’s coming next. Instead of just extracting training data from gamers and building models that might replace them, Nerve creates a pathway for gamers to participate in the AI economy.
It’s a smart business move too. As General Intuition’s models get more capable, they’ll need human oversight for edge cases, safety validation, and real-world data collection. Having a ready-made workforce that already understands spatial environments — because they’ve been navigating virtual ones for years — is a competitive advantage that’s hard to replicate.
The ethics line
De Witte spent three years working in humanitarian spaces, including with Doctors Without Borders, and has drawn a clear boundary: no agents will be employed to harm humans. The company explicitly prohibits military applications, though de Witte says he’s open to search-and-rescue use cases.
This matters because the same technology that can navigate a robot through an office could theoretically navigate one through a battlefield. As Silicon Valley grows increasingly comfortable with defense contracts, General Intuition’s stance is notable — especially backed by investors like Khosla and Bezos who have their own positions on defense tech.
The company’s chief of staff, Brianna Martin, publicly quit Palantir over its work with ICE. De Witte brought her on specifically because of that decision. It’s a signal about the kind of company General Intuition wants to be, not just what it wants to build.
What this means for the AI agent space
If you’re building with AI tools — whether it’s automating your business or creating content with AI — General Intuition’s approach suggests the next wave of AI agents won’t come from bigger language models. They’ll come from models that understand the physical world.
The API launch this summer will be the real test. Can third-party developers build useful applications on top of General Intuition’s world model? Can the model hold up across diverse real-world environments, not just a controlled office demo? The company has a handful of customers in gaming, simulation, and robotics today, but the ecosystem play is what justifies a $2.3 billion valuation.
De Witte’s vision is to be an ecosystem enabler — a model provider that lets others build on top of its technology. “We’re not gonna build a self-driving car company,” he said. “We’re gonna make it 10 times easier for the next person to build a self-driving car company.” That’s the same playbook that worked for OpenAI with language, applied to spatial intelligence.
The bottom line
General Intuition is no longer a fascinating experiment. With $454 million in total funding, a working robot demo, a jobs platform, and an API on the way, it’s a real company with real products. The question isn’t whether video game data can train AI agents — they’ve demonstrated that. The question is whether the approach scales to the messy, unpredictable physical world at a level that justifies the valuation.
For anyone watching the AI space, this is one to track. The API launch this summer will tell us a lot. If you want to stay on top of what’s actually working in AI tools, I cover the developments that matter — not just the ones that make headlines. And if you’re new to AI agents, start here — I break down the concepts without the jargon.