If large language models are the brains of machines, then dexterous robotic hands are becoming their bodies. As AI steps out of the digital world and into the physical one, Linkerbot has chosen a tougher and more fundamental route: building robotic hands capable of handling the real world with human-like precision.
“NVIDIA CEO Jensen Huang once said the next wave of AI would be physical AI,” the Linkerbot team said. “To us, the real bottleneck isn’t just computing power (brain). It’s how quickly the hardware (body) can catch up.”Only when robots truly learn how to see, touch, move and adapt on their own can embodied intelligence begin to take shape, the team added.
From building dexterous hands to powering embodied intelligence
Over the past few years, Linkerbot has focused on developing high-DOF (Degree of Freedom) dexterous robotic hands, gradually building a full-stack system that combines hardware, algorithms, and data. Now, the company is moving beyond pure hardware manufacturing toward a new stage of intelligent evolution.
Linkerbot has introduced a PEEK-based (Polyether Ether Ketone) reducer, replacing traditional metal structures to make robots lighter and more efficient. The company also explores multiple technical paths — tendon-driven, linkage-based, and direct-drive — while developing AI-powered dexterous hands with integrated perception.
On the software side, Linkerbot is building a physical world model and has launched its large-scale model, the Linker Creation Model, along with Linker OS and a real-world reinforcement learning framework, aiming to close the loop from data collection to model training.
“Our goal is not just to build robotic hands, but to create the foundational infrastructure for embodied intelligence,” the team said.
Three key scenarios driving real-world demand
For real-world deployment, Linkerbot is currently focusing on three core application areas: precision manufacturing, healthcare and life sciences, and home services.
In factories, its dexterous robotic hands are already being used for micro-component assembly in the automotive and consumer electronics sectors. In agriculture, the company is working with domestic research institutions to develop breeding robots and explore the use of dexterous manipulation in high-tech farming. In eldercare and rehabilitation, dexterous robotic hands are being developed to support caregiving and daily tasks — from folding clothes to pouring coffee, much like a family member would.
“Our goal is to free people from dangerous, physically demanding, and repetitive work, so they can focus on more creative tasks,” the team said.
Industrial-grade reliability is the biggest hurdle
Compared with lab demos, the toughest challenge in real-world deployment is earning trust in the physical world, according to the team.“In the lab, a bug is just a software patch,” the team said. “But in real-life scenarios, a bug can mean physical damage or even safety risks.” That, they added, is why dexterous robotic hands have long been seen as fragile and unreliable.
With this in mind, Linkerbot aims to strike a balance between durability and cost. Its Linker Hand series has survived tough tests, from crashes to drop tests, and has been involved in setting national standards. On the global ManipulationNet benchmark, Linkerbot has completed remote Peg-in-Hole manipulation validation, a key test for fine motor control.
Meanwhile, the company has launched a modular teleoperation system called Open TeleDex. Built around the TripleAny concept — any hand, any arm, any device — it makes system integration much simpler and more flexible for customers.
Bridging academic research and industrial deployment
Overseas, Linkerbot mainly focuses on research communities in North America and Europe, and its products are being used in labs at universities such as Stanford and Cambridge. At the same time, the company is also paying close attention to healthcare and eldercare needs in aging societies such as Japan and South Korea.
“Technology knows no borders, but culture still matters,” the team said. Many overseas developers care deeply about open-source tools and interoperability, so Open TeleDex is built with an open architecture. This allows developers to control robotic hands using familiar devices such as VR headsets and exoskeletons, making it easier to experiment, build, and collaborate with others around the world.
The Linkerbot team said their customers generally come to them with two very different types of needs. One group is focused on pushing the limits of physical-world dexterity, mainly for research and exploration purposes. The other is more concerned with building reliable and stable systems that can be deployed in real industrial scenarios and scaled for mass production.
Scaling up is the real turning point for embodied intelligence
When it comes to manufacturing, Linkerbot can now produce over 1,000 high-dexterity robotic hands per month, and is aiming to scale that up to 50,000–100,000 units per year by 2026. In December 2025, Linkerbot closed its Series A++ funding round led by Sequoia China and CCV (Creation Partners Capital).
The Linkerbot team believes that scale is where real impact begins. “When we deliver 100,000 dexterous hands, that means 100,000 dangerous, exhausting, or repetitive jobs can be taken over by robots,” the team said. “That’s our long-term commitment to using technology for good.”
At a time when embodied intelligence is still in its early days, Linkerbot is betting that more human-like hands, produced at real industrial scale, may lay the groundwork for the future of physical AI.
