Physical intelligence emerges as next frontier beyond LLMs, humanoids
NEW YORK, UNITED STATES — As the technology sector remains fixated on large language models (LLMs) and humanoid robotics, a fundamental shift toward physical intelligence is gaining urgency.
Driven by unsustainable labor demands in logistics and manufacturing, a new class of purpose-built machines is emerging to master the complexities of the real world, with companies like Ambi Robotics, GrayMatter Robotics, and autonomous vehicle firms leading the charge.
Fortune notes, “The next frontier isn’t digital intelligence that can describe the world. It’s physical intelligence that can change it.”
The high-stakes shift from digital to physical intelligence
The current artificial intelligence (AI) boom is built on what analysts describe as “2D AI,” which uses LLMs trained on trillions of internet-derived text tokens.
However, this approach fails to translate to the physical world, where machines must master physics, gravity, and consequence without the luxury of a pre-existing “physical internet” to scrape for training data.
In the 2D world, an AI hallucination is a typo; in the 3D world, it is a physical failure—a robot crushing a parcel, tipping a pallet, or crashing a vehicle. Unlike digital errors, physical failures carry immediate, tangible consequences, requiring the development of “world models” that simulate cause-and-effect through interaction.
This change requires an entirely new way of learning, reflective of the way humans acquire skills by processing data produced through interaction rather than through text.
The amount of data necessary to enable a machine to execute a simple task, like walking across a room, is exponentially larger than the amount of data gathered from Shakespeare’s works, which provides a substantial strategic moat.
Mastering this new frontier involves not just understanding space but also the dimension of time, where AI must simulate future outcomes—predicting if a box will slip three seconds from now—before executing an action in logistics, manufacturing, or defense.
Rethinking automation beyond humanoid form
As much effort is being channeled into creating general-purpose humanoid robots, Fortune reports that focusing on this area is one of the major misinterpretations of industrial evolution.
Humans are made to hunt and gather as opposed to working hours on the hardship of heavy boxes and breathing dust in the sanding booths of factories.
Building machines that replicate the human body’s physical limitations, such as legs for sorting packages or humanoid forms for sanding jet parts, ignores the opportunity to create superior, specialized tools.
The next generation of industrial automation will hence be purpose-built machines that perform a particular task, leaving human workers without jobs that are economically and morally unsustainable.
Examples of such companies include Ambi Robotics, which deploys suction-based robotic arms to perform heavy tasks in warehouses, and GrayMatter Robotics, which is focused on automating potentially dangerous surface-finishing tasks to eliminate people from dust clouds.
Equally, self-driving car companies such as Stack AV and Waymo are confronting the grueling realities of long-haul trucking, not regarding these technologies as job killers but as body saviors that protect human health and redirect their work toward creativity and judgment.
The report concludes, “The winners won’t be those who build the most convincing chatbots. They’ll be those who build the nervous system for the physical world. It’s time for AI to leave the screen and enter the warehouse, the factory, and the street.”
“That’s where the real world — and the real value — resides.”
As physical AI moves from chatbots to task-specific machines in warehouses, factories, and transport, the future of work is likely to be defined less by human replacement than by a deeper shift of labor away from dangerous, repetitive tasks and toward oversight, problem-solving, and higher-value roles.
