1X CEO Bernt Børnich on NEO: Why Safe Humanoid Robots Learn Faster

1X founder and CEO Bernt Børnich explains why the NEO humanoid robot is built around safety, tendons and an open developer platform, and why he expects 2027 to be the year of physical AI. In this 13-minute talk at Modal’s Runtime 2026 (published November 2, 2026), NEO walks on stage to introduce him. Børnich then argues that humanoids will become infrastructure the way GPUs did, that a robot which is not safe cannot learn, and that developers, not 1X alone, will build most of what NEO ends up doing. He says NEO ships to developers and some consumers in December.

TLDW

Børnich lays out 1X’s three founding principles: robots must be safe enough to live and learn among people, as capable as humans, and cheap enough to manufacture at scale. He argues that NEO’s tendon drive, loosely modeled on muscle, makes it lighter, safer and more capable than geared robots, and calls hands “the final boss of robotics.” The second half covers the NEO developer platform: data collection rigs that mirror the robot’s sensors, simulation and training tools, an automated data pipeline, open APIs, and community efforts like a weekly leaderboard streamed on Twitch.

Thoughts

Børnich’s best idea is a reframing of safety: a robot that is not safe cannot learn. The usual argument is commercial. A robot has to be safe to be allowed into homes, hospitals and shops. His point goes further. If these systems learn from real-world experience, they will only be “as intelligent as their experiences,” and the diversity of that experience is capped by where the robot is allowed to go and what it is allowed to get wrong. A caged factory arm never sees a cluttered kitchen. A heavy, geared robot cannot afford trial and error near people, so it never collects the failure data that reinforcement learning in the real world needs. That turns safety from a compliance cost into a data strategy, and it is a sharper explanation than most robotics companies give for why their hardware choices matter to their AI.

The developer platform may matter more than the robot. 1X is shipping data collection rigs that copy NEO’s cameras and tactile sensors and track the whole body, so a person can record demonstrations that retarget directly onto the robot. That data, along with teleoperation sessions and the robot’s own policy rollouts, feeds an automated pipeline that segments tasks, writes dense captions and runs filtering and QA before training. Børnich compares 1X to Nvidia, with a consumer front end and a developer back end, and the comparison fits. The scarce input in robotics is diverse, labeled, real-world data, and a mass-produced rig that turns any developer into a data source is a way to grow that supply without 1X paying for every hour of it. Open APIs that let people swap in their own tools are a sensible admission that a small team will not build the best version of every piece.

The boldest claim is that NEO breaks the usual trade-off: lighter and safer, but also more capable. Børnich knows how that sounds (“keep me honest”) and frames it with an analogy: classic geared robots are gasoline cars, and a tendon-driven robot is an electric car that works on different principles. The physics argument is reasonable. With no fast-spinning gears, the robot carries little kinetic energy while still applying real force, and putting the actuators away from the hand mirrors how human forearms drive the fingers. The proof will come from the December shipments, not the stage. Hands are where the claim will be tested hardest, since he admits that the range of tasks a robot can do, and therefore the diversity of its data, comes down to dexterity and touch.

The community plans at the end say a lot about where 1X thinks the bottleneck is. Børnich calls the field “fragmented,” with no shared place where everyone builds together, and proposes Friday NEO Live: a leaderboard where the top three community-trained policies each week run on a real NEO, streamed on Twitch. Add hackathons at 1X offices for people who cannot afford a robot, and it looks like a play to become the default open platform for embodied AI research, the way a common hardware target helped earlier waves of software. His closing ambition, robots staffing wet labs and building particle accelerators rather than just cleaning kitchens, is far off. But it explains why a consumer humanoid company is spending its stage time courting researchers.

Key Takeaways

Børnich expects little to change in physical AI through 2026, but thinks 2027 will be a breakout year.
1X’s mission is a world of abundance where humanity is no longer limited by labor, and Børnich thinks physical AI will matter even more than digital AI for that goal.
He sees humanoids becoming infrastructure like GPUs, providing labor, data collection and learning signal from the real world.
1X has three founding principles from day one, 11 years ago: safe, as capable as a human, and affordable at scale.
Humanoids only make economic sense at volume, like phones or computers, which are complex but cheap and reliable at scale.
Safety is a learning requirement: a robot that cannot safely go somewhere cannot collect data from it.
NEO uses a proprietary tendon drive, from materials and motors up to manufacturing and models, so it can be strong without carrying much kinetic energy.
Compliant hardware makes mistakes cheap, whether that is not crushing a carrot or retrying a dropped warehouse box.
NEO ships in December 2026 to developers and some consumers.
Hands are “the final boss of robotics,” and Børnich does not think the problem can be solved without tendons.
The developer platform includes body-tracking data rigs with NEO’s cameras and tactile sensors, simulation and training tools, and open APIs.
1X built an automated data pipeline in the last six months that segments tasks, writes dense captions and handles filtering and QA.
1X plans a Discord community, research partners in industry and academia, hackathons, and a weekly Twitch-streamed leaderboard for community policies.

Chapters

0:08 NEO Takes the Stage

NEO walks out and introduces itself as the one non-human intelligence in the room that is not on a screen or in a sandbox, then frames the talk as being about production when the model has a body. It hands over to its “creator,” Bernt Børnich.

0:59 Why 2027 Will Be the Year of Physical AI

Børnich predicts a quiet 2026 and a dramatic 2027 for physical AI. He states 1X’s mission of labor abundance and argues the physical side of AI will play an even bigger role than the digital side.

1:51 Humanoids as Infrastructure for Developers

Humanoids will become infrastructure like GPUs, he says, supplying labor and real-world learning signal. Like Nvidia, 1X has a consumer front end and a developer back end, and the back end will be built by the developer community.

3:45 Safe, Capable, Affordable: 1X’s Founding Principles

Raised on science fiction and expecting to end up in a Japanese robotics lab, Børnich joined 1X 11 years ago. The company’s principles have not changed: robots safe enough to live among people, capable enough not to be toys, and simple enough to manufacture at scale.

5:00 Safety as a Requirement for Learning

Caged factory robots have limited applications. More importantly, robots that learn in the real world are only as good as the diversity of their experiences, and an unsafe robot cannot go where that diversity lives or learn from trial and error the way people do.

5:50 Tendons vs. Gears: Inside NEO’s Drive

NEO’s tendon drive, loosely inspired by muscle, lets it be strong without fast-spinning gears and the kinetic energy they carry. Børnich claims this breaks the “no free lunch” rule, making NEO more capable as well as lighter and safer, and compares geared robots to gasoline cars and NEO to an electric car.

7:37 NEO Ships in December, and Hands Are the Final Boss

NEO ships in December to developers and some consumers. Hands decide how many tasks a robot can do, and therefore how diverse its data is. A decade of tendon research pays off here, since even nature puts most of the hand’s muscles in the forearm.

8:21 The NEO Developer Platform

The platform includes mass-produced data rigs that mirror NEO’s sensors, the robot itself, simulation and training tools, and an automated pipeline for segmentation, captioning, filtering and QA. Data from rigs, teleoperation and policy rollouts all feeds in, and APIs let developers plug in their own tools.

10:50 Discord, Friday NEO Live and Hackathons

Calling the field fragmented, Børnich invites developers into a Discord community launching with industry and academic research partners. Friday NEO Live will run the top three community policies on a real NEO each week on Twitch, and hackathons will let people without a robot deploy code.

12:17 Beyond the Kitchen: Labor Abundance

Børnich closes on automation well beyond chores: building the largest particle accelerators and staffing wet labs to tackle open questions in bioscience and science in general.

Notable Quotes

“Most of the non-human intelligence in this room lives on screens or in sandboxes. I am the exception. I walked in.”
NEO, introducing the talk

“I think ultimately these machines are going to become infrastructure just like GPUs.”
Bernt Børnich, on the future of humanoid robots

“They need to be as capable as us, as dextrous, as strong, and as agile. If not, they’re just toys.”
Bernt Børnich, on 1X’s founding principles

“If they’re not safe, they can’t learn.”
Bernt Børnich, on why safety matters beyond market access

“I like to think about like the classical robotics more like gasoline cars and NEO is more like an electric car.”
Bernt Børnich, on tendon drives versus gears

“The final boss of robotics is hands, right?”
Bernt Børnich, on dexterity and data diversity

Watch the full talk here.