Two months ago, most people had never heard of Generalist. Now it's worth $3 billion, and the round isn't even fully closed.
That kind of trajectory used to be reserved for chatbot startups riding the generative AI wave. Now it's happening in robotics, and Generalist is the latest proof that venture capital has found its next obsession: teaching machines to physically do things, not just talk about them.
According to two people with knowledge of the funding cited by TechCrunch, Generalist has raised nearly $200 million in additional capital led by 8VC, pushing its valuation to $3 billion. A regulatory filing confirms the size of the new capital. This isn't a fresh, standalone round. It's an extension of the $400 million Series B that Generalist announced in June, which was led by Radical Ventures and valued the company at $2 billion at the time. With this extension, the total round now stands at $600 million.
Business Insider reported the same $3 billion figure while the round was still in discussions, also naming 8VC as the expected lead. Neither Generalist nor 8VC responded to requests for comment from either outlet, so the exact terms remain unconfirmed. But two independent reports landing on the same number, from different investors' mouths, is about as close to solid as pre-close funding news gets.
Do the math on the timeline and it's the part that actually stops you: a 50% jump in valuation in roughly two months, without the company even bothering to formally close and announce a new round. Investors aren't waiting for a clean quarterly cadence. They're chasing the deal before someone else gets there first.
Generalist was founded in 2024, which makes this whole valuation run even more remarkable. The founders aren't first-timers guessing at robotics. Pete Florence and Andy Zeng came from Google DeepMind's robotics research group, and Andrew Barry spent time as an engineer at Boston Dynamics, arguably the most recognizable name in advanced robotics hardware. That pedigree explains why early money came in fast: 8VC and Radical Ventures backed the company early, alongside Nvidia, Union Square Ventures, Bezos Expeditions, and AI researcher Fei-Fei Li. Business Insider also reported angel backing from Xiaomi cofounder Bin Lin and AngelList cofounder Naval Ravikant.
Until recently, Generalist operated quietly, without much public presence. That's notably different from the AI industry's usual playbook of loud launches and demo videos designed to go viral. It suggests a company that was building conviction among insiders long before it needed the outside world to pay attention.
Generalist isn't making robots. It's making the AI model that tells robots what to do, an approach sometimes described as building the "brain" that can sit inside different robotic bodies rather than one company's proprietary hardware.
Its earlier model, GEN-1, was built to teach robots basic physical tasks, and the company released videos showing robots folding laundry, sorting Lego bricks, and using a whisk to mix baking ingredients. Its newer release, Gen 1.5, pushes further: according to TechCrunch, it claims to enable robots to learn new tasks from video demonstrations as short as 3 to 12 seconds. If that holds up outside of curated demo footage, it's a meaningful jump from the old approach of exhaustively programming or training a robot for each specific task.
The company is reportedly working with a handful of customers already, using their feedback to adapt the model for specific use cases. That's a small, unglamorous detail, but it matters. It means Generalist isn't just chasing valuation headlines, it has real deployments generating real feedback loops, however early.
Generalist is far from alone. Physical Intelligence is reportedly valued at more than $11 billion. Skild AI, backed by SoftBank, is valued at more than $14 billion. Genesis AI was reportedly in talks last month to raise at a $3 billion valuation of its own, and Field AI sits around $2 billion. That's four to five well-funded companies all racing toward the same idea: a foundation model general enough to control a wide range of robots without task-specific retraining every time.
The bet driving all this money is that robotics is approaching its own "ChatGPT moment," the point where robots handle general tasks the way large language models handle general text, without needing to be explicitly trained on each one. It's an appealing thesis, and it's clearly moving billions of dollars right now.
But there's a real catch, and it's worth taking seriously: robots can't be trained on the entirety of the internet's data the way LLMs can. Text is everywhere. Video and language are scrapeable at massive scale. Physical, embodied interaction data is not, it has to be generated slowly, expensively
META BLOCK