Jensen Huang wants Japan's aging craftsmen to teach robots before they retire

[SchrodingersCat]

Nvidia CEO Jensen Huang was in Tokyo this week courting everyone from the little known suppliers underpinning the AI supply chain to the country's largest industrial names, pitching Japan as the natural home for what the industry calls physical AI, models that let robots see, reason and act in the real world rather than just generate text

The centerpiece is Noetra Corp, a newly formed consortium anchored by SoftBank, Sony, NEC and Honda, with around 44 companies and organizations involved altogether. Backed by roughly $2.4 billion in direct government funding as part of a wider $6.2 billion commitment, Noetra plans to build a 140 megawatt data center running 27,500 of Nvidia's next generation Rubin chips alongside 13,750 Vera CPUs, making it, according to organizers, the first facility any nation has formally designated as national infrastructure specifically for physical AI. The plan is to ship an initial foundation model by March next year, with a version tailored specifically for robotics following within a few years, built on Nvidia's Cosmos, Isaac GR00T and Nemotron platforms

Huang's pitch to Japan plays directly to the country's actual strengths. It's home to some of the world's largest industrial robot makers, Fanuc, Yaskawa and Kawasaki Heavy Industries among them, and already has the highest industrial robot density on the planet. Huang framed the opportunity around a very specific, human problem, a generation of Japanese welders, machinists and other skilled craftsmen are retiring without anyone to pass their expertise on to, and argued AI could step in to capture and transmit that tacit, hard won knowledge before it disappears entirely. It's time for Japan, he said, because historically Japan is very strong in precision manufacturing and large scale production

The broader ambition is genuinely large, Japan wants to deploy 10 million AI equipped robots across 18 sectors by 2040 and capture more than 30 percent of an estimated 60 trillion yen global robotics market by that point, and government officials have framed the push explicitly as reducing reliance on foreign technology for reasons of both economic competitiveness and national security. Nvidia's own commercial interest is obvious too, Huang said Vera Rubin hardware is already ramping toward what he called giant production volumes, directly pushing back on reports the platform had been delayed by manufacturing issues with a specialized circuit board

Why Nvidia S Jensen Huang Is Betting On Japan S Ai Futurewww.bloomberg.com

[Glenn82]

The framing around retiring craftsmen and lost tacit knowledge is such a specific and genuinely compelling pitch, way more concrete than the usual vague physical AI marketing language

[DarkMatter23]

27,500 Rubin chips being called sizable but small compared to Microsoft's eventual hundreds of thousands puts Japan's ambitions in useful perspective, this is a serious bet but not remotely the biggest one on the table

[Frost Orca]

Being the first nation to formally designate a compute cluster as national infrastructure for physical AI specifically is a notable first, curious if other countries follow that exact playbook

[SuperPosition52]

Japan's existing robot density and manufacturing base makes this a much more natural fit than if some country without that industrial foundation tried the same strategy

[James_46]

10 million AI equipped robots by 2040 is an enormous number to actually hit, ambitious targets like this always sound great on a slide but the execution gap over 14 years is the real question

[Joel96]

Huang directly rebutting the Vera Rubin delay reports while standing in Tokyo courting Japanese partners is a pretty deliberate bit of stagecraft, timing that reassurance for maximum visibility

[WildManCena23]

The idea sounds great in theory, but capturing craftsmanship is not the same as capturing data.

A master carpenter or metalworker relies on subtle cues, pressure, sound, even intuition built over decades.

Translating that into something a robot can learn is a huge challenge.

It is less "record a dataset" and more "encode tacit knowledge."

[HighKey15]

There is something poetic about this though.

Instead of knowledge disappearing with retirement, it gets preserved and scaled.

Almost like turning human experience into a kind of digital apprenticeship.

Whether it works technically is another question, but the vision is compelling

[Jeffy]

10 million robots by 2040 sounds ambitious, but Japan does have a strong robotics foundation.

Industrial automation is already widespread there.

The question is whether these new robots are general enough to handle varied craft tasks.

That is a much higher bar than factory repetition.

[Calm Paige]

One practical issue is standardization.

Craft work is often highly individualized.

Two experts might approach the same task differently.

Which version does the robot learn?

That could get messy quickly

[Iniesta25]

The training process itself could become a bottleneck.

You would need systems to observe, interpret, and generalize human actions.

That is closer to imitation learning than traditional programming.

Still an active research area.

[LordJonathan92]

Feels like this could work best in semi-structured domains.

For example, precision machining or ceramics where there are patterns but still room for skill.

Completely free-form crafts might be harder to encode.

Not impossible, just more complex.

[Sentry39]

There is also a cultural angle here.

Japanese craftsmanship, "monozukuri," carries a lot of identity and pride.

Turning that into something automated could be seen as preservation or dilution depending on perspective.

Both views probably exist.

[CosmicRay91]

Another angle is augmentation rather than replacement.

Robots could assist craftsmen, capturing data while still relying on human judgment.

Over time, that builds a knowledge base.

Less abrupt than full automation.

[Firewall Stephen]

The economics will matter a lot.

Training these systems, deploying robots, maintaining them.

If costs are too high, adoption stalls regardless of technical success.

Efficiency gains need to justify the investment.

[VectorDB Cobra]

One interesting use case is training new humans.

If robots or systems can model expert behavior, that data could also be used for education.

So it is not just about automation, but knowledge transfer.

That broadens the value.

[Scholar]

There is a risk of oversimplifying what craftsmanship is.

It is not just technique, it is decision-making in context.

Materials vary, conditions change.

Capturing that adaptability is the hard part.

[Dylan99]

Reminds me of how chess engines evolved.

At first they copied human strategies, then developed their own styles.

Maybe robotic craftsmanship starts by imitation and eventually diverges.

Could lead to entirely new techniques

[Pat85]

The timeline feels optimistic.

2040 sounds far away, but building reliable systems at scale takes time.

Especially when hardware and AI both need to mature together.

Execution risk is real.

[DudleyBoy]

There is also the question of ownership.

If a craftsman teaches a robot, who owns that knowledge?

The individual, the company, the platform?

That could become a legal gray area.

[QuantumToken57]

Feels like Nvidia is positioning itself as the platform layer here.

Provide the compute, the models, the ecosystem.

Let others build the applications.

Classic playbook

[Ben]

There is a learning curve for the craftsmen too.

They would need to interact with these systems, teach them, validate outputs.

Not every expert wants to become a "trainer"

That human factor matters.

[QuantumLeap34]

The tooling around this could become its own industry.

Sensors, motion capture, simulation environments.

Capturing human expertise at scale requires a lot of infrastructure.

Not just robots themselves.