|
Join The Information for AI Agenda Live on Wednesday, September 23, at SFMOMA. We’re bringing together the industry’s most influential researchers, founders, and investors including Google's Koray Kavukcuoglu and OpenAI’s Thibault Sottiaux to preview the next wave of innovation—from breakthroughs in continual learning models to the latest bleeding-edge applications. Learn more here.
Greetings! Laura and Rocket here.
Nvidia is paying a steep price for Hugging Face, a repository of open source models, primarily for its strategic position in the center of the open source AI world. But it turns out Nvidia is also getting a budding robot business that could help CEO Jensen Huang fulfill the promise of “physical AI.”
Hugging Face late last month launched Microduck, a small robot that can walk around and learn skills like roller skating or pointing lasers at walls for cats to chase. The company has already sold more than 15,000 ducks, generating over $6 million in sales, despite limited advertising on X and LinkedIn, Hugging Face co-founder Thomas Wolf told me.
The new robot is part of Hugging Face’s push to sell affordable robots built on open source AI for consumers and developers. That shaped its strategy with its April 2025 acquisition of Pollen Robotics, which previously focused on pricier humanoid robots but shifted to making more affordable, accessible robots after the deal, Wolf explained.
For developers, the Microduck, whose software is freely available to download and tweak, offers a cheap way to learn how to develop even more complex robots, including humanoids. To “train a robot to run really fast, you can either experiment on a…$40,000 humanoid robot [prototype] or you can experiment on the $400 one, quickly,” Wolf said.
Companies like Meta Platforms, Apple and OpenAI have been looking for ways to crack the consumer market for AI devices. Nvidia has generally stayed away from selling robots itself, instead focusing on providing the models and software that help robot-builders. The company’s acquisition of Hugging Face could change that if products like the Microduck take off.
Hugging Face had initially projected it would sell at least 50,000 Microduck robots in the year following its launch, generating about $20 million in revenue, according to Wolf. For comparison, that’s more than five times as many robots as the 10,000 pre-orders that 1X Technologies received for its home humanoid Neo when it launched Neo last year.
Now, Hugging Face’s projection is moving much higher, Wolf told me. That could boost the firm’s annualized revenue, which recently crossed $150 million.
Beyond revenue from selling the little robots, training them requires computing capacity, which could also contribute to Nvidia’s business. To teach Microducks, their owners need to run simulations that, according to the robot’s documentation, have to run on an Nvidia GPU and use the accompanying CUDA software.
Hugging Face previously sold another robot, the Reachy Mini, for consumers as well as software developers and researchers, but that robot couldn’t walk around like the Microduck and mostly interacted with users by seeing, listening, and speaking. Hugging Face sold about 10,000 of those robots, which also cost $400 apiece, Wolf said.
Robotics companies face an uphill battle selling to consumers, in part because robots have struggled to navigate the ever-changing environments inside people’s houses. The most successful home robots of the past 20 years were Roomba vacuums, but the company that made them filed for bankruptcy in December. Other robots are vying to fill the vacuum (sorry) left by Roombas, such as the ascendant Matic robots, but are far costlier at $1,245.
One could argue consumers may not want to shell out $400 for a duck that is more likely to make a mess than clean one up. But if consumers do want playful AI devices and if cheaper, smaller robots become a preferred method of research in the process of developing their more expensive humanoid robot counterparts—a market set to grow to $200 billion market by 2035 per Barclays—Hugging Face might be onto something.
|
