Some venture investors are betting that robotics is approaching its own “ChatGPT moment,” when machines can perform a wide range of tasks without being trained separately for each one. The latest sign is Generalist, a two-year-old startup developing a single AI model for different kinds of robots. According to two people familiar with the terms who spoke to TechCrunch, the company has reached a $3 billion valuation after raising close to $200 million in an extension led by 8VC.

The timing matters because the capital is being deployed before the underlying science has fully caught up, underscoring how much of the robotics bet is now focused on software that can translate general intelligence into physical action. Axios, which first reported the financing, said money is “flying fast” into companies building “robot brains,” or software designed to help machines interpret and act within the world around them.

Why the round is an extension, not a fresh raise

The new capital is not a separate financing round. Instead, it is an extension of Generalist’s $400 million Series B announced in June, when Radical Ventures participated at a $2 billion valuation. The latest extension lifts the total amount raised in the round to $600 million.

The financing is confirmed by a federal Form D filing. The filing shows $208.2 million offered and $198.2 million sold to 32 investors. The issuer is listed as Generalist AI, Inc., a Delaware company based on Market Street in San Francisco and registered in 2024. The company was formerly known as Artificial General Dexterity, Inc.

The documents name executives including Peter Florence, Andrew Barry and Andy Zeng, along with directors including Ellen Chisa and Fraser Kelton. Axios reported that Generalist declined to make a statement, but the financing details were disclosed through the filing documents.

Founded by former DeepMind and Boston Dynamics employees

Generalist was founded in 2024 by Florence and Zeng, who previously worked as researchers at Google DeepMind, and Barry, who previously worked in engineering at Boston Dynamics. In June, the company said it was moving toward what it called “physical AGI” and said it had already raised more than $500 million.

TechCrunch reported that Generalist’s backers include 8VC, Radical Ventures, Nvidia, Union Square Ventures, Bezos Expeditions and AI researcher Fei-Fei Li. In June, 8VC described Generalist as a “frontier lab for robot intelligence” and identified Florence as CEO, Zeng as chief scientist and Barry as CTO.

What Generalist says its model can do

Generalist does not build robots themselves. Instead, it develops the software that controls them. The company says its Gen 1.5 model can learn from video demonstrations that are 3 to 12 seconds long, and that it is continuing development with a small group of customers, according to TechCrunch.

In a July research post, Generalist said its earlier GEN-1 model worked across roughly 9,000 variations of robot hands and grippers and was trained on more than half a million hours of interaction data.

Generalist still trails the two highest-valued rivals cited by TechCrunch, while Genesis AI was reportedly seeking the same $3 billion valuation.

Investors remain cautious about the science

Despite the surge of funding, the core technical challenge remains unresolved. Unlike large language models, robots cannot simply be trained on the entire internet, and some venture investors say a truly general-purpose robot model may still be years away. That gap helps explain why so much of the current investment is flowing into foundation-model startups rather than full robot makers: the software layer is where backers see the chance to standardize capabilities across many machines, even if the field has not yet proven that approach at scale.

That caution is not limited to Silicon Valley. At the 2026 World Robot Conference in Beijing, experts said embodied AI trails large language models (LLMs) by at least five years, according to Yicai. Zhang Jianzhong, chairman and chief executive of chipmaker Moore Threads, said most vision-language-action models have about seven billion parameters, far below the hundred-billion scale LLMs reached before their breakthrough. He also pointed to a shortage of real-world data as a major constraint.

“ It is certainly still some way from its ‘GPT moment.'” — Zhang Jianzhong, chairman and chief executive of Moore Threads, cited by Yicai Global

TechCrunch said investors are effectively wagering on a robotics version of a “ChatGPT moment,” meaning general-purpose models that can transfer skills across different robots and tasks.

The flow of capital is extending beyond AI model developers. Cryptopolitan previously reported that Xpeng raised more than $900 million for its robotics arm at a $6.3 billion valuation in a round led by IDG Capital. Together, the deals show how aggressively investors are positioning for a robotics breakthrough, even though the general-purpose “robot brain” they are backing has not yet been proven.