AI Has Finally Stepped Off the Screen

BPJ β | Experimental Voyage

Semiconductors connected by light, AI employees with login IDs, camera-readable cards — thinking about a printing company’s next adventure

As I used AI to follow news and academic papers, four stories that looked unrelated connected into one line.

In Japan, massive investment is beginning in factories designed to connect AI-compute semiconductors (GPUs and the like) to each other using “light.”

To make those semiconductors, production will ramp up for enormous manufacturing equipment that draws microscopic circuits with light.

At banks overseas, AI is starting to get login IDs and receive assignments from human managers.

And a technology was published that lets an ordinary camera recognize trading cards and overlay 3D-style effects onto a livestream.

Semiconductors, banking, card games.

At first glance, they seem to have nothing to do with each other.

But to me, they all looked like the same shift.

AI is no longer something that lives only inside a screen.

It is starting to build factories, run machines, do work inside organizations, recognize physical products and cards, and even change the experiences people have.

That is interesting.

And for printing companies—who have long worked at the boundary between digital and physical—there may be more of a role here than we think.

This time, while tracking the latest AI news, I want to do some slightly bold imagining about what kinds of work printing companies could create next.

NEWS FRONT | Four stories became one line

1. Japan will get a “factory that connects AI with light”

On July 14, 2026, semiconductor maker Tower Semiconductor announced it would invest about $3 billion in Japan to expand production capacity. The Japanese government also plans to support the effort with about $1 billion. $3 billion is about ¥450 billion (approx.) if $1 = ¥150. $1 billion is about ¥150 billion (approx.)—a noticeably large class of domestic-scale factory investment.

Reuters | Tower Semiconductor to invest $3 billion in Japan

At the center is a technology called “silicon photonics.”

It is a difficult name, but in plain terms it means: semiconductor technology that lets devices exchange data not only through electrical wiring, but also through light.

As the amount of information AI handles grows, what matters is not just the compute power of each chip, but how fast—and with how little power—you can connect chips to one another.

I understand this as a technology for building light-speed highways inside AI’s gigantic brain.

Along with retrofitting existing fabs, a new 300 mm wafer-capable fab is also planned, and the first equipment is aiming for full-scale operation around the end of 2027. “300 mm wafer” refers to the diameter of the round substrate used to make semiconductors—the mainstream size for mass production. “Full-scale operation” is easiest to picture as the stage where you are no longer just prototyping; you are running stable, repeatable mass production.

The “capacity expansion” here also, roughly speaking, means “increasing how many wafers and how much process volume the factory can handle over a given period.”

AI’s growth is starting to move not only software companies, but factories, equipment, talent, regions, and logistics inside Japan.

2. The “light-based printing presses” that make semiconductors will ramp up too

The next day, July 15, semiconductor equipment giant ASML significantly raised its 2026 sales outlook and indicated a plan to increase production capacity for its most advanced manufacturing equipment by about 30% each year over the next two years.

Reuters | ASML capacity upgrade soothes AI chip bottleneck fears

ASML is globally critical for “lithography” exposure tools that draw semiconductor circuits. The cutting edge is the area called EUV lithography, and it directly links to ramping up AI-chip production.

What ASML makes is equipment that uses extremely short-wavelength light to transfer ultra-fine circuit patterns onto semiconductors.

“Transfer” here is a little different from transferring ink onto paper. It is closer to the idea of using light to bake a circuit “mold” into a photosensitive material, then in later steps removing and leaving material to form the circuit.

It is a different field, but as a printing company, you cannot help reacting to the shared idea of an “enormous machine that draws patterns with light for mass production.”

Also, “about 30% each year” is reported not so much as a sales story, but as a story about increasing how many tools ASML can produce and deliver. Still, from the article alone, it is hard to tell whether this is the number of units itself, or shipping capacity including parts supply.

Either way, the world right now is not only short of chips to run AI.

Even the machines needed to make those chips are starting to have their future production slots filled.

AI is shifting from “a convenient app” into a huge industry that pulls in factories, equipment, materials, electricity, and engineers.

3. AI got an employee ID—and a manager

On July 13, Reuters reported on how major US and European banks are starting to embed AI into everyday operations.

Reuters | Wall Street banks ramp up digital assistants

The most striking example was BNY (Bank of New York Mellon).

The company’s “digital employees” are not just a metaphor. They are designed as work AIs with internal accounts. They have login IDs, are assigned responsibilities, and human managers check the quality of their work. They are not simply chat systems that answer questions; they enter internal systems, do tasks, and leave records of activity.

The term “AI agents” here also refers to AIs that are not instruction-waiting chatbots, but can assemble a procedure and push tasks forward with some independence.

The report also introduces survey data saying 51% of banks are piloting these kinds of AI agents. “Piloting” here is not just a lab PoC; it is closer to trying something in a limited department or workflow in a way that resembles real operations.

“AI employees” used to be a futuristic metaphor.

Now it is becoming a very practical organizational design question—AI with IDs, responsibilities, managers, evaluations, and permissions.

Bunseikaku’s own “AI employee” concept also needs to move beyond simply giving an AI a name. We have to decide what that AI reads, what it drafts, what it can execute, and who is responsible.

4. An ordinary camera turns trading cards into a digital experience

A research project released on July 2 called “TCG-AR” was also fascinating. TCG stands for trading card game, and AR stands for Augmented Reality.

arXiv | TCG-AR: Real-Time Multi-View Augmented Reality for Trading Card Game Streaming

Note that arXiv is a public site where many manuscripts are posted before journal publication (preprints). Interesting seeds show up early, but the burden of judging reliability still sits with the reader.

The system uses an ordinary camera to distinguish trading cards during gameplay, recognize a card’s orientation and type, and overlay 3D-style effects and game information onto the livestream video. It aims to run without special chips or a dedicated playmat—using consumer cameras and general streaming software. It is easiest to imagine software commonly used for streaming, like OBS.

In other words, the printed card itself becomes the entrance that summons a digital production.

Place the card.

The camera recognizes it.

A character moves on-screen.

For viewers, the current game state is displayed clearly.

Livestreaming card battles—previously flat—starts to feel closer to sports broadcasting or game streaming.

Instead of viewing trading cards as “printed matter,” think of them as an interface that connects real-world tactility with digital effects.

This is where Bunseikaku’s paperboard printing, cards, IP goods, events, and interest in AI connect neatly. “IP goods” here means officially licensed merchandise that uses works and characters—anime, games, and other rights-managed properties.

AI is no longer just a software story

Putting these four stories side by side changes the outline of AI.

AI is not only something that generates text and images inside a chat window.

It requires semiconductors connected by light, requires enormous equipment to manufacture those semiconductors, gets produced in factories, receives assignments like an employee inside companies, identifies physical cards and products, and changes human experiences.

In other words, AI is stepping into the real world.

Masayoshi Son of SoftBank also said in June that he is advancing a “Physical AI” factory where robots build robots. If you take “Physical AI” here to mean the domain where AI operates real machines and robots and delivers outcomes as on-site work, it connects to this thread. In a Reuters survey of Japanese companies, one in three said they have already introduced AI robots, plan to, or are considering them—and the main use case was manufacturing floors.

Reuters | SoftBank's Son on the physical AI plant

Reuters | One in three Japan firms using or considering AI robots

There is still exaggeration and expectation-leading-the-way in parts of this.

Not every job will suddenly be replaced by AI.

Even so, the direction is becoming quite clear: AI’s main battlefield is expanding from inside the screen into factories, logistics, retail floors, events, and offices.

And watching this change, I started to think that printing companies are not outside AI. We might be standing exactly at the place where AI connects to the real world.

Why printing companies have a role here

Printing companies do not handle data alone.

We turn data into paper, cards, packaging, booklets, labels, POP, and signs. (POP refers to in-store promotional materials meant to drive purchase.)

We produce thousands of identical pieces accurately.

We change content for just one piece.

We control color and shape.

We inspect.

We insert and pack.

We sort by store, by customer.

We deliver to a specified place on a specified date.

From the perspective of digital companies, the real world is quite troublesome.

It has weight.

It has dimensions.

Materials vary.

Once something is printed, it cannot be easily revised.

Shipping takes time, and if you make a mistake, you may have to retrieve physical goods.

Printing companies have dealt with that troublesome reality for decades.

That is why, the more AI steps into the real world, the more a printer’s abilities—“the power to make things accurate,” “the power to deliver without mistakes,” “the power to operate on-site”—start to take on different value.

AI can create a concept.

But the work of converting that concept into the right size, color, material, and quantity—and into a form that can actually be used on-site—will likely remain.

AI can organize information.

But making it understandable to engineers, customers, local communities, and job candidates still requires editing, design, printing, video, and experience design.

AI can recognize cards.

At the same time, whether that card is beautifully printed, easy to identify, and desirable to fans depends on quality on the physical side.

That is where I see an interesting future for printing companies.

Because if AI and semiconductor factories get built, it will not only be equipment and buildings—they will also need huge amounts of information.

Five new kinds of work that seem likely to be born next

1. Build everything a new factory needs to “communicate”

If AI and semiconductor factories get built, they will need massive amounts of information—not only equipment and buildings.

Recruiting materials.

Safety training.

Work procedures.

Factory tour guides.

Technical explanations for partners.

Explanations for local residents.

Trade show panels.

Sales booklets and videos.

Rather than making each item separately, you interview engineers, organize the information, and deploy it across paper, the web, video, and in-factory signage.

This is work where a printing company builds a factory’s “communication system.”

It also feels quite realistic as something Bunseikaku could offer manufacturing firms in Ota City (Tokyo’s well-known small-factory district).

2. Create manuals that are easy for AI to use

Until now, manuals were made on the assumption that humans read them.

From here, more work will likely appear where you produce both a paper manual and AI-searchable data “from the same primary source.”

Break procedures into fine steps.

Tie photos and diagrams to specific process steps.

Organize common questions and cautions.

Make update dates and owners explicit.

On-site, people can check with a card or booklet—and ask an AI when they do not understand something.

A very practical layer of work emerges between printing and AI.

3. Make cards and packaging that cameras can read

QR codes are not the only way to connect paper and digital.

As image recognition advances, cameras can recognize card artwork, packaging shape, and the label itself—calling up video, AR, games, audio, and product information.

At a trading card event, show a card to a camera and the character moves.

On a factory tour, hold up an equipment-explanation card and a video begins.

Point a camera at product packaging and the producer’s story and usage instructions appear.

Many of these are still in the experimental phase, but printed matter can shift from “something you read” to “something that responds.”

4. Design the work of AI employees

Deciding what AI employees should do is also a new domain.

Organize the content of RFQs (requests for quotation).

Notify sales of missing information.

Search past jobs.

Collect sources for articles.

Draft task lists after meetings.

But:

Do not send anything to customers on its own.

Order acceptance confirmation and price decisions are done by humans.

Do not send customer information outside.

You compile these roles, permissions, and review procedures onto a single sheet. It is essentially a job description: a document that defines “what it is allowed to do, and where humans must take responsibility.”

We could start by experimenting inside Bunseikaku, and if we find a form that is genuinely useful, perhaps we can share it with other small and mid-sized firms and printing companies facing the same questions.

5. Turn physical media into “membership cards” and “keys to experiences”

The more AI generates huge volumes of images and text, the more the meaning of owning something physical may strengthen.

Limited-edition cards.

Membership cards.

Booklets with serial numbers.

Event participation kits.

Content only purchasers can access.

Physical letters delivered from artists and creators.

Paper and cards cannot beat digital in raw information volume.

But they can offer the feeling of ownership, commemorative value, trust, scarcity, and the pleasure of sharing with others.

Do not treat printed matter as a digital substitute. Treat it as a “key to experiences” whose value rises precisely because it is the digital age.

In music, publishing, anime, games, trading cards, and community-building, there still seem to be many possibilities.

If we were to test something at Bunseikaku starting this week

Before making any big investment decisions, there are small things we can test.

The first is an experiment in card recognition.

Prepare about 20 types of cards printed at Bunseikaku (or test cards), and use a smartphone camera to see how stable recognition remains when you change angle, lighting, reflections, varnish, and background.

Even just studying the relationship between print quality and AI recognition seems like it would yield interesting findings.

The second is creating a single-sheet job description for an AI employee.

For example, an “RFQ cleanup AI.” This is an AI whose role is to format RFQ text, identify missing items, and make the next confirmation step easier.

What it reads.

What it is allowed to draft.

What it must never execute.

Who reviews it.

The metric you measure one week later.

Decide these, then compare humans and AI on just ten cases.

The third is an experiment where you pick one factory technology and expand it into five formats.

Interview an engineer for 30 minutes.

From that, create: a one-sheet sales handout, a factory-tour card, a web article, recruiting copy, and a short video script.

With AI, the first pass of organizing and adapting can go quite fast.

Then on-site staff review and correct it into accurate information.

If this works, it could grow into a new service for manufacturing DX (transformation that includes both digitizing the shop floor and changing how work itself is done).

Deliberately, let’s also consider the opposite possibility

If AI spreads, paper decreases.

In a sense, I think that is true.

Confirmation documents and paper that does not need to be stored will keep decreasing.

But it is not guaranteed that everything becomes digital.

The more AI-generated information increases, the more important it becomes to know what is authentic, who issued it, and which experiences were shared.

That is where physical cards, booklets, certificates, packaging, and signs may take on new roles.

Printed matter may stop being a container for large volumes of information.

Instead, it may become something that signals trust.

Makes ownership tangible.

Gathers people.

Starts a digital experience.

Leaves a memory.

Those roles could grow stronger.

The more AI grows, the more the value of meaningful printed matter rises.

It is still a hypothesis, but I want to keep chasing this direction.

Of course, this is not an easy story

If you let AI employees use internal systems, you need permissions and information governance.

If you want cameras to recognize cards, you have to think about reflections, damage, lighting, and misrecognition caused by similar designs.

If you connect paper to digital, you also need to decide “who will manage what, and for how many years” to prevent dead links.

If you feed factory information to AI, you must verify that old procedures and incorrect information are not mixed in.

Just because a technology is interesting does not mean it becomes a business overnight.

That is exactly why Bunseikaku does not want to “pretend we understand.” We want to try small tests on the shop floor and learn—including from failures.

BPJ will also turn those experiments into a record.

Printing can give AI a body

AI writes.

It makes images.

It calculates.

It searches for information.

Those capabilities will only grow.

But to reach people in the real world, it needs a body.

Factories.

Machines.

Robots.

Cards.

Packaging.

Booklets.

Signs.

Events.

These are the contact points where people pick something up, look, move, and remember.

Printing companies do not only compete with AI.

We can turn AI-generated information into forms that can be used in the real world.

We can build entrances that go from physical objects back into AI and digital experiences.

We can convert difficult technology into words and forms people can understand.

Seen that way, printing’s future is still very interesting.

A factory that connects AI with light.

AI with an employee ID.

Trading cards that a camera can read.

These are not strange stories from distant industries.

They may be nautical charts arriving from beyond the horizon—maps for printing companies to start their next adventure.

We at Bunseikaku will start with small experiments.

Collect with AI.

Test on the floor.

Connect paper and digital.

And create jobs that do not yet have names.

That is how we think about Beyond Printing.