Can Japan Win Back the “Brain” of Robotics?

A next course for printing companies—seen through 27,500 GPUs and made-in-Japan physical AI

Collected with AI, thought through on the shop floor, and turned into words about the future of print.

BPJ β | Experimental Voyage

This morning, while I had an AI scan global headlines, I stumbled on a story that felt genuinely big.

In Japan, a serious push is now underway to build a “domestic AI brain” for robots—the kind of core intelligence that can run machines in the real world.

At the center is a new company called Noetra, with participants including Sony, SoftBank, NEC, and Honda.

Noetra plans to build a large-scale AI compute platform by lining up roughly 27,500 of NVIDIA’s latest GPUs, and then develop AI models that Japan’s manufacturing sector and robot makers can actually use.

In a Reuters interview dated July 22, 2026, Noetra’s CEO said this effort could be Japan’s “last chance,” speaking with a sense of crisis.

That’s strong language.

But there are reasons behind it.

As the US and China race ahead in AI development, Japan still has world-class robot makers, manufacturing equipment, sensors, materials, and factories.

Japan can build the machines.

It also has deep shop-floor knowledge.

What’s been missing is the AI “brain” that can understand all of that, learn from it, and drive it.

What Noetra is aiming for is not just “a ChatGPT that’s good at Japanese.”

It watches factory video.

It understands how people and machines move.

It recognizes space and the position of objects.

It tells robots what to do next.

In other words: AI that understands the physical world.

That’s… pretty interesting.

And when you think about it, printing companies are not that far from this world.

Every day, printers turn data into physical reality.

They combine paper, ink, machines, color, temperature, humidity, and human judgment to reproduce the same quality—thousands or tens of thousands of times.

As AI starts to move beyond the screen, the “shop-floor data” inside printing plants may begin to carry a different kind of value than it did before.

A quick look at this week’s printing industry

First, a simple read on where the printing industry stands right now.

Print demand isn’t moving in one single direction

According to the latest monthly report from the Japan Printing Industries Federation, Japan’s printing production value in April 2026 was ¥31.1 billion (about $200 million, approximate) up 2.8% year-on-year.

Packaging printing rose 5.0%, and commercial printing rose 1.0%.

Meanwhile, publishing printing fell 3.3%.

Rather than the entire print market shrinking uniformly, it seems closer to saying that demand is shifting between use cases.

Paper volumes keep falling

Shipments of printing paper fell 7.1% year-on-year, marking the 18th straight month below the prior year.

Markets built on printing huge volumes of the same thing remain tough.

That’s exactly why value has to be created not in copies, but in planning, finishing, operations, and even the end-user experience.

The direction of capex is changing

Production units of printing machinery were down 14.3% year-on-year, the ninth consecutive month of decline.

At the same time, governments and large corporations are starting massive investments in AI, semiconductors, robots, and data centers.

For printing companies, the next investment target isn’t only printing presses.

How do you connect existing equipment through data?

How do you preserve human judgment?

How do you turn customer back-and-forth into a system?

We’re entering an era of investing before and after the press—upstream and downstream—not just in the machine itself.

Even if sales rebound, profit and talent are the bottlenecks

In JAGAT’s FY2025 “Printing Industry Management Capability Survey,” the 99 respondent companies reported sales increasing for the second consecutive year.

At the same time, profits were thin, and the average employee age was rising.

That said, the survey reports that training spend per employee increased, and there are signs of improvement in productivity indicators.

In other words, the management theme for printing companies is shifting from

not only

to

What “physical AI” means: AI gets a body

What Noetra is developing is an AI known as a “multimodal foundation model.”

“Multimodal” means it can handle multiple types of information at once—not just text, but images, video, audio, and more.

Beyond that, by FY2030, it aims to realize “Real-world Native AI” that understands space and the properties of objects, and can operate in the real world.

It sounds complicated, but put simply:

.

Noetra has investment from 44 companies and organizations.

Development participants also include the National Institute of Advanced Industrial Science and Technology (AIST) and Preferred Networks.

Construction of the AI compute platform is slated to start in April 2027, with operations planned for June 2028.

According to NVIDIA, the platform is planned to use roughly 27,500 Rubin GPUs and about 13,750 Vera CPUs, forming an AI infrastructure on the scale of 140 megawatts.

This is not only a plan to build AI models.

It is also a plan to create an environment where on-site data held by Japan’s factories, logistics operations, robots, automotive sector, and telecoms can be used safely within Japan.

Japan’s Ministry of Economy, Trade and Industry (METI) also states that Japan has broad shop-floor data across manufacturing, and that leveraging it for physical AI will strengthen industrial competitiveness.

Printing plants also have “data AI still doesn’t know”

So what kinds of data do printing companies have?

Even for a single print job, for example:

• Paper type and lot
• Ink and varnish conditions
• Temperature and humidity
• Press speed
• Color measurement values
• Registration adjustment
• Waste (spoilage) sheets
• Reasons the machine stopped
• Defect images
• Adjustments made by the operator
• The final judgment that the quality passed

A lot of this information is scattered across forms, daily reports, machine logs, and people’s memories.

Because it isn’t consolidated as data, it can’t be reused across the company.

But what if it were organized?

Find similar past jobs.

Recommend appropriate initial settings.

Predict how many rounds of color adjustment will be needed.

Detect defect images.

Spot signs of breakdowns or consumable wear.

Explain expert judgment to new hires.

Rebuild the production schedule.

AI could potentially support this kind of work.

Of course, that doesn’t mean you can fully auto-run a press overnight.

Print quality involves judgments that can’t be explained with numbers alone.

There are customer-specific preferences.

And there are many exceptions—specialty stocks, finishing, inks, climate, and more.

That’s why the first thing you need is not expensive robots.

You need to convert what the shop floor is judging into a form that can be recorded.

Printing companies aren’t only “users” of physical AI

For many small and mid-sized firms, 27,500 GPUs is a world away.

There’s no need to develop your own foundation model.

However, once physical AI spreads, the shop-floor data that AI learns from will come from companies that actually operate factories.

That matters.

AI companies have the technology to build AI.

Manufacturers and printing companies have the knowledge that has made real-world work succeed.

Under what conditions does quality stabilize?

What sound is the early warning sign of failure?

Which defects are acceptable, and which require stopping the line?

What quality points do customers truly care about?

This knowledge does not exist in sufficient quantity on the internet.

If Japan wants to compete in physical AI, small and mid-sized factory-owning companies may be able to become not just AI users, but teachers of the real world.

But to do that, they must protect their data, organize it, and decide who to share what with—and how far.

This is not a story of “just hand everything over” to an AI company.

Data will become one of the factory’s most important assets.

New work for printing companies, born from the AI/robot wave

As physical AI spreads, new demand will emerge not only inside printing companies, but on the customer side as well.

When new robots and AI devices launch, customers will need:

• Product packaging
• Instruction manuals
• Setup cards
• Safety signage
• In-store displays
• Trade show panels
• Training materials
• Maintenance and inspection sheets
• On-site factory signage
• Case study booklets

But it won’t end with making paper.

Hands-on demo events where people can actually touch the product.

Tours of factories and showrooms.

Registration of buyers and users.

How-to videos on the web.

AI chat support for inquiries.

Post-use follow-ups and update notices.

Work that connects print, experience, and online into a single flow will increase.

This is where Bunseikaku’s “Beyond Printing” research can lead to something real.

Instead of finishing at “we delivered the printed matter,”

think together from

It’s work that connects future products with ordinary people.

The roles of working people will change, too

In the era of physical AI, how we see work inside printing companies will change.

A press operator isn’t just someone who runs a machine.

They are someone who judges quality, tunes conditions, and leaves that knowledge for the next production run.

Sales isn’t just someone who quotes price and delivery.

They are someone who understands the customer’s new technology and business, and translates it into paper, experience, and web.

IT isn’t just someone who fixes PC problems.

They are someone who understands shop-floor work and turns it into data and systems.

At the boundaries between printing, manufacturing, AI, and systems, job roles that didn’t exist before will appear.

The printing industry is not only an industry that protects old technology.

It can become an industry that connects real-world handling technology with digital systems.

For younger people, it should be a genuinely interesting career.

If you want to start small next week

Seeing “27,500 GPUs” doesn’t mean you have to conclude, “That has nothing to do with us.”

A first experiment can be just one machine, one product, and one week.

Build “process data you can teach to AI”

Pick one standard repeat job. For 20 production runs, record the following:

• Paper
• Quantity
• Press speed
• Temperature and humidity
• Number of adjustments
• Reasons for stoppages
• Waste sheets count
• Color measurement values
• Defect images
• Operator judgment
• Final result

The goal is not to let AI operate the machine immediately.

When you line the data up, the goal is to find:

• What judgments aren’t being recorded?
• Which items are written differently depending on who’s filling it out?
• Which conditions seem likely to relate to quality?
• Which information is worth preserving next time too?

This small dataset becomes the foundation for B-DOCK, AI employees, quality control, training, and automation.

Before making big investments, confirm the shop-floor knowledge you already have.

Start there.

Turn “the last chance” into our story

Noetra’s CEO said it might be Japan’s last chance.

It’s true: looking at global AI investment and development speed, Japan isn’t in a position where it can simply catch up.

Noetra’s plan also has challenges.

It requires enormous power.

Dependence on NVIDIA products remains.

It’s unclear whether government funding will reach real businesses and regional shop floors.

And we have to think carefully about how AI-driven automation will affect human work.

Even so, Japan has manufacturing sites it can be proud of on a world stage.

Printing companies, too, have shop floors that have dealt with paper, ink, color, machines, and human judgment.

We are not companies that build AI models.

But we can understand our own shop floors through data.

We can use AI to raise quality.

We can connect customers’ new products across print, experience, and online.

Rather than watching the giant wave of physical AI from far away, we can pull it closer to our own work.

Printing companies aren’t outside the future.

We’re in the place where data becomes physical objects—and physical objects generate data.

As Japan tries to win back the brain of robotics, printing companies may also be at the moment to re-examine the knowledge embedded in their shop floors.

Use AI to understand the world.

Turn shop-floor strengths into data.

Test small.

And create jobs that don’t yet have names.

The printing industry is setting out on another interesting voyage.

Reference information

• Reuters | Robot AI company Noetra is “last chance” for Japan, CEO says
• Noetra / Sony / SoftBank / NEC / Honda | Full-scale launch toward developing a domestically produced multimodal foundation model
• NVIDIA | Japan Government, Industrial Leaders and NVIDIA Launch the World’s First National AI Infrastructure
• Ministry of Economy, Trade and Industry (METI) | Multimodal foundation model development project with AI robots / physical AI in view
• Sony | Exhibited research-and-development aibo at SIGGRAPH 2026
• Japan Printing Industries Federation | Printing Industry Monthly Report, June 2026 issue
• JAGAT | JAGAT info, July 2026 issue

#BeyondPrinting #BeyondPrintingJournal #BPJ #printing #printingcompanies #physicalAI #robots #manufacturingDX #AI #factoryDX #qualitycontrol #skillstransfer #monozukuri(manufacturing) #Bunseikaku