Meta Platforms has completed the acquisition of startup Assured Robot Intelligence (ARI), fully entering the race for humanoid robotics. A company spokesperson confirmed the transaction, but the financial terms were not disclosed.
ARI was no ordinary startup. The company focused on developing foundation models for humanoid robots—systems designed to enable machines to understand human behavior, predict it, and adapt to complex, changing environments. In other words, not merely a robot that repeats programmed movements, but a machine that learns.
The project was founded by two scientists with impressive résumés. Xiaolong Wang was a researcher at Nvidia and served as an associate professor at the University of California, San Diego. Lerrel Pinto taught at NYU and previously co-founded Fauna Robotics, a startup focused on human-sized robots for children. Amazon liked the project so much that it acquired it in March 2026. Pinto is therefore joining Meta almost immediately after selling his previous company. The entire ARI team, including both founders, will join the Meta Superintelligence Labs research division.
Zuckerberg has been thinking about robots for years
The company’s research teams have been working on robotics technologies for several years, and internal reports leaked last spring described ambitions to build a humanoid robot directly for consumers, including proprietary AI models and hardware.
ARI could help move this plan forward. According to a Meta spokesperson, Pinto and Wang’s team brings deep expertise in designing models for robot control and autonomous learning, specifically for full-body control of a humanoid robot. This is an area where most companies are still trying to find their way.
The acquisition comes two days after Meta announced an increase in its planned capital expenditures for this year. The company raised its estimate by $10 billion to a new range of $125 billion to $145 billion, citing higher component prices and the cost of artificial intelligence data centers.
The path to artificial general intelligence runs through physical AI
Why would a company that makes money from advertising and social media invest in robots capable of doing household chores?
Many AI experts today believe that the path to artificial general intelligence—the theoretical point at which AI matches or surpasses human capabilities in every field—does not lead solely through the processing of textual data. Models must learn in the physical world through direct interaction with their environment. A robot that falls and gets back up, or picks up a glass without breaking it, must solve problems of a different kind than a language model generating answers from text. And this is precisely the gap that ARI will help fill.
The race for humanoid robots is shifting into high gear
Meta is not alone. Amazon acquired Fauna Robotics. Agility Robotics is testing robots in Amazon warehouses. Figure AI, Physical Intelligence, Apptronik, and other startups are raising hundreds of millions of dollars. And then there is Tesla with Optimus.
The robotics market is attracting investor attention, although estimates of its size vary dramatically. Goldman Sachs estimates that the market will reach approximately $38 billion by 2035. Investment bank Morgan Stanley even forecasts $5 trillion by 2050. Such a wide range illustrates just how uncertain the sector is.
Meanwhile, Meta is pushing ahead on several fronts simultaneously. In April 2026, it laid off approximately 10 percent of its employees, citing efficiency. China blocked its attempted $2 billion acquisition of AI startup Manus. And Zuckerberg acknowledged during an internal meeting that an armed conflict had partly caused the slowdown in revenue growth and that the layoffs were a direct consequence of AI costs.
Nevertheless, in April the company unveiled the first model from its new Superintelligence Labs group, called Muse Spark, which Meta says competes with models from Google, OpenAI, and Anthropic.
Sources: wsj.com, techcrunch.com, and bloomberg.com
