1X plans to ramp its humanoid output in quality-gated steps, with a Hayward plant ramping to about 10,000 units and a San Carlos plant targeting 110,000 units a year at full capacity.
Bernt Børnich argues production is far harder than prototyping because rare faults only surface at larger volumes, and 1X is aiming for 50,000 robots by the end of next year to feed its data loop.
Børnich says 1X can solve the general robot problem rather than automating tasks one by one, arguing a general model can be distilled into specialised applications but not the reverse.
1X bets simple design and general AI can scale humanoids
1X is betting that humanoid robots become a mass product through simple hardware and one general AI model rather than task-by-task automation. A design with about 1,000 parts and in-house motors allows fast iteration and volume, while deployed robots supply the data that improves the model. The target of 50,000 robots next year is meant to start that loop, with safety as the barrier to clear.
Takeaways
Production, not prototyping, is the hard part. The Hayward factory is built out for about 10,000 units at full ramp, and a San Carlos factory for about 110,000 a year is mostly due late next year. Rare faults only appear at volume, so 1X ramps in steps through quality gates.
A general model beats task-by-task automation. Børnich argues 1X can solve the general problem. A general model can be distilled into specialised applications but not the reverse, and robot data is diversity bound rather than data bound. He expects specialised models to win in 2026 and general intelligence to shift in 2027.
Deployed robots feed the learning loop. Robots in the field generate successes and failures that improve the models, which is why 1X targets 50,000 robots out by the end of next year. Børnich wants models that learn how the world works rather than clone human behaviour.
Simple hardware enables fast iteration at scale. A well designed humanoid has about 1,000 parts against roughly 50,000 in a car. Tendon drives with in-house motors remove complex gearing. The key metric is about four weeks from a major CAD change to a new robot off the line.
Safety is the main barrier to trust. 1X hopes to prove safety more formally this year, and a safe humanoid should weigh under 70 pounds. Trust builds slowly because each hardware change needs new data from few homes, such as the Eve pilot in Børnich's own home.
Beautiful products come from simple designs and small teams. Børnich uses a cable and connector budget and cites a motor shaft that lost one part. He would keep the team smaller for longer, and he sees the most valuable engineers as those who are broad across domains.
Warmth and play drive human acceptance. 1X designs robots with warmth rather than fear so the technology blends into society. Robots learn social behaviour from general models, and delight is largely a data problem that remains unsolved. Børnich says play matters for learning.
Plans for this year
Enterprise deployments with very large customers are planned alongside home robots.
The NEO platform will open to developers, because 1X cannot solve robotics alone.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Folding is not the main application, and Eve folded shirts years ago.
The longer-term dream is household chores on demand.