2025-08-15

Robotics CEO: Humanoid Robots Are Coming to Your Home This Decade w/ Bernt Bornich & Dave Blundin

Peter Diamandis and Dave Blundin tour 1X Technologies in Palo Alto and interview founder/CEO Bernt Bornich about the Neo Gamma home humanoid robot. Bornich explains why 1X bet on the home instead of factories (data diversity beats repetitive factory tasks for building toward AGI), walks through Neo Gamma's specs, price point (~$30k or ~$300/month), safety and privacy architecture, and the blend of teleoperation and full autonomy. The conversation covers world models as a safety-evaluation tool, the manufacturing and chip-fab bottlenecks to scaling to billions of robots, and why China's hardware ecosystem and process knowledge outpace the US. It closes on 1X's hiring pitch, pre-orders landing in 2025, and a vision of humanoids ushering in material abundance and helping solve hard science problems.

▶ Watch on YouTube

Topics

Why 1X bet on the home, not factories Robotics ▶ 4:00

Bornich argues consumer hardware scales faster than anything else, and that intelligence comes from data diversity, which the home offers and repetitive factory tasks don't.

Data diversity as the bottleneck to robot intelligence AI ▶ 5:01

1X's prior robot Eve plateaued in learning after 20-40 hours on narrow factory/guarding tasks; home deployment removes that cap on task diversity.

Neo Gamma price point and ownership model Robotics ▶ 13:41

Bornich confirms a price point near Diamandis's guess of ~$30k purchase or ~$300/month lease, competitive with cheaper Chinese robots despite higher capability.

The humanoid as companion Robotics ▶ 15:30

Bornich frames the robot as something between a pet and a human, a lifelong companion that remembers everything about its owner, calling it 'the hobbes' (Calvin and Hobbes).

Why intelligence isn't language-first AI ▶ 17:49

Bornich argues core intelligence is spatial/temporal, not linguistic, and that models trained natively on vision and action (plus text) will beat language-first models.

Teleoperation blended with autonomy Robotics ▶ 37:38

Neo Gamma mixes 'best-effort autonomy' with human teleoperators giving high-level guidance; the two meet in the middle as learned autonomy expands and teleop recedes.

Privacy and safety architecture for an in-home robot Robotics ▶ 43:35

No human at 1X can see raw data without owner approval and a decryption key; a 24-hour training delay lets owners delete anything before it enters training weights.

World models for safety evaluation AI ▶ 58:17

1X trains world models that simulate outcomes of robot actions, letting them test new models for performance and safety before deploying to real robots, since you can't A/B test physical behavior on customers.

Manufacturing scale and the labor/materials bottleneck Robotics ▶ 23:29

1X targets 20k+ robots/year by end of 2026 and hundreds of thousands/year by decade's end, but warns that aluminum refining and labor (not just demand) will throttle the ramp, as it did for the iPhone.

Why the humanoid form wins (for now) Robotics ▶ 1:01:56

Answering Salim Ismail's 'why not six arms' objection, Bornich says humanoid form transfers human skill data most easily and is the most general shape, comparing it to the mainframe-to-PC phase of computing before specialized robot forms emerge.

China's hardware ecosystem advantage Geopolitics ▶ 1:08:21

Bornich credits China's edge to deep disseminated manufacturing process knowledge (e.g., magnet production) and free economic zones with fast permitting, more than top-down subsidy alone.

Venture capital's hardware aversion Economy ▶ 1:13:59

Blundin and Bornich discuss how US venture capital avoided hardware/physical-component startups for years, hurting returns, while the biggest companies today are all hardware-based.

Chip fab and GPU bottleneck for scaling robots Compute ▶ 1:35:47

Diamandis flags that each robot may need a full GPU or two, only ~20M GPUs are made per year, and TSMC holds 66% fab market share, making chip capacity a hard ceiling on humanoid scaling.

Humanoids, abundance, and hard science Robotics ▶ 1:24:00

Bornich's 10-year vision: abundance of labor and energy enables sustainable abundance for everyone, and humanoids are necessary to staff the particle accelerators and biotech labs needed to solve remaining hard science problems.

Predictions made

open Peter Diamandis: Peter Diamandis will have his own Neo Gamma robot at home by the end of the year.

· due: 2025 · ▶ watch

“I just secured my first Neo Gamma at my home by end of the year.”

Your call:

open Bernt Bornich: 1X's factory will run at a rate north of 20,000 Neo Gamma robots per year by the end of 2026.

· due: 2026 · ▶ watch

“the factory run rate end of 2026 is north of 20k”

Your call:

open Bernt Bornich: 1X will reach a production run rate of hundreds of thousands of robots per year before the end of this decade.

· due: 2029 · ▶ watch

“you can imagine a run rate before the end of this decade of hundreds of thousands per year”

Your call:

open Bernt Bornich: 1X will open pre-orders and take orders for Neo Gamma within 2025.

· due: 2025 · ▶ watch

“It's this year.”

Your call:

open Bernt Bornich: There will be roughly 10 billion humanoid robots on Earth by 2040, possibly sooner, depending on supply-chain and permitting constraints.

· due: 2040 · ▶ watch

“I think it's probably roughly correct. I think it might happen before.”

Your call:

open Bernt Bornich: 1X's world-model technology will eventually be productized and sold as its own capability, beyond just powering Neo Gamma.

· due: unspecified · ▶ watch

“yes, this will be productized... still a bit out”

Your call:

open Bernt Bornich: Neo Gamma will eventually be allowed to cook and handle other currently-restricted tasks once 1X has enough certainty that the AI's behavior is safe.

· due: unspecified · ▶ watch

“over time, as the AI improves and we get more and more certainty on all behaviors being safe, we will allow cooking”

Your call:

open Bernt Bornich: After an initial general-purpose humanoid phase, robotics will fragment into specialized robot forms built for specific tasks, mirroring how computing specialized after the PC era.

· due: unspecified · ▶ watch

“we will get to where we have Star Wars, there will be different drones doing different tasks”

Your call:

Numbers that matter

1X's Eve robot plateaued in learning after 20-40 hours on a given narrow task Deployed in guarding/logistics 2022-2023; more varied tasks (e.g. guarding with wheeled navigation) land near 40+ hours, repetitive tasks near 20.
Neo Gamma price point approximately $30k to purchase, or ~$300/month to lease (~$10/day, ~40 cents/hour) Bornich confirmed Diamandis's guessed numbers as roughly right, calling it already price-competitive with cheaper, less capable Chinese robots.
10,000 robots generate more non-duplicated useful data per day than is uploaded to YouTube daily Bornich's napkin-math argument for why robot fleets will out-produce the internet as a data source.
1X's factory run rate will exceed 20,000 robots/year by end of 2026 Ramping toward hundreds of thousands/year by end of decade, modeled loosely on the iPhone's roughly 1.7x annualized growth to reach a billion units.
1X robot has a few hundred parts vs. a car's ~50,000 parts; the robot weighs 66 lbs vs. a car's ~4,000 lbs Bornich argues humanoids are closer in complexity to a refrigerator than a car.
Neo Gamma: 66 lbs, lifts ~150 lbs, carries ~50 lbs, ~4 hour battery life, recharges in 30 min to 2 hours, stands 5'4" Given as the core spec sheet; Bornich notes weight-to-strength ratio matches an athletic human.
Neo Gamma has roughly 42 degrees of freedom: 7 per arm, 3 in the spine, 3 in the neck, 22 in each hand Compared to the ~26 DOF common in other humanoid robots, which often substitute a neck for wrists.
Global GDP is about $110 trillion, with labor roughly half of that, implying a $50+ trillion addressable market for robot labor Cited as the scale of opportunity if humanoids simply replicate existing human labor, before any new categories of work.
1X's custom motors produce 5.5x the world record in torque Enables tendon-driven actuation instead of gears, which Bornich says drives the robot's light weight, compliance, and low manufacturing cost.
Roughly 20 million GPUs are manufactured per year; TSMC holds about 66% of chip fab market share; each robot may need 1-2 GPUs Diamandis's argument that GPU/fab capacity, not robot demand, will be the binding constraint on reaching billions of humanoids.
For comparison, there are roughly 1 billion automobiles and about 8 billion smartphones on the planet today, against a 10 billion humanoid robot target by 2040 Used to illustrate how far short current chip/GPU production is from supporting that scale.

Worth digging into

🕳️ 1X's published white paper on world models and safety

Bornich directly cites it as the technical case for why 1X bets heavily on world models over other approaches, including for automated safety evaluation and red-teaming.

🕳️ The rare-earth magnet and process-knowledge bottleneck

Bornich and Blundin both flag that China's near-monopoly isn't just raw materials but decades of disseminated manufacturing process knowledge (e.g., magnet production), a bottleneck that's been openly discussed for years without Western action.

🕳️ Chip fab capacity vs. the 10-billion-robot target

Diamandis's math (20M GPUs/year, TSMC's 66% share, 1-2 GPUs per robot) suggests the 2040 projection for 10 billion humanoids is implausible without a major fab buildout.

🕳️ 1X's world-model-driven safety/evaluation pipeline vs. competitors

Bornich claims 1X is roughly three months ahead of competitors on world models and treats this as a core defensible moat tied to the industry's monster valuations.

🕳️ The venture capital hardware drought and its reversal

Blundin and Bornich claim VC avoided hardware for years, hurting fund returns, and that this is only now reversing as hardware-based companies dominate the biggest outcomes.