To double its shipbuilding, Japan is 60,000 workers short. Can wall-climbing robots and AI that learns from master craftsmen close the gap?

You've probably seen footage of a car plant: rows of robot arms throwing sparks as they weld car bodies.

Now picture a shipyard. Even on a ship the size of a building, most of the seams between steel plates are still welded by human hands.

Japan has now decided to build about twice as many of these handmade ships by 2035. But the number of people working in its shipyards keeps falling. If you can't double the workforce, the only option is robots and AI, and ships are among the hardest things for robots to build.

How are Japanese shipyards tackling that contradiction? DATA WORLD went through government documents, company announcements and trade press and listed every case since 2022 of a Japanese shipyard bringing in, testing or planning robots or AI. We found 72. We also read 28 tech job postings from shipyards and compared Japan with yards in Korea, China and the US.

Robots thrive in car plants because identical bodies keep coming down the line. Teach a robot once where and in what order to weld, and it can repeat that tens of thousands of times.

Ships are different. A single ship has more than 100,000 parts, and each customer's design is different, so you never get hundreds of identical hulls in a row. A 2003 Japan Welding Society survey, now dated, found that robots did about three-quarters of the welding in car and other transport equipment plants, while shipbuilding automated only about a third, even counting simple welding carriages. In 2024 Kawasaki Heavy Industries wrote that most shipbuilding processes are still done by hand.

Why do robots struggle so much? Shipyard and government documents point to four reasons.

• Every ship is one-off. Japan's transport ministry (MLIT) notes that because each ship has its own design, it can't be mass-produced on a line.
• Huge, and never exact. Large steel parts shift a little every time they are assembled. Kobe Steel cites this as a reason robots miss their target weld position.
• Tight, high, curved. Ships have double bottoms with cramped, enclosed space between them. The outside is high up and the plates at bow and stern curve. Government documents say welding and painting in tight spaces still need human hands.
• Teaching robots takes time. Before a robot can weld, a person has to teach it where and how. Namura Shipbuilding says this teaching makes up half of all automatic welding work.

That last point is easy to miss. If every hull is different, the teaching starts over for every hull. However fast the robot welds, if people still spend their time preparing it, the workforce doesn't shrink.

Flip that around: if AI can cut the teaching, shipbuilding changes fast. That is the key to this whole story.

Japan now wants twice as many of these handmade ships. The December 2025 Shipbuilding Revitalization Roadmap sets a 2035 target of 18 million gross tons a year (gross tonnage measures a ship's size), roughly double today. Scaled from 2024's 390 ships, that is about 770 ships a year.

The money has started to flow. In September, plans at eight yards, including Imabari, JMU, Mitsubishi Shipbuilding and Kawasaki Heavy Industries, were approved: about ¥900 billion (roughly $6 billion) of investment, with up to about ¥311 billion of state support.

This isn't the first such goal. A 2020 government advisory report set almost the same figure, 18.3 million tons, as the target for 2025. Actual 2025 output was 10.1 million. The summit has simply moved ten years further out.

Twice the ships needs twice the people

Why the miss? The roadmap itself names a shortage of people in design and on the shop floor as one cause.

Japanese skilled workers who weld and assemble steel in the yards fell from about 71,000 in 2010 to about 47,000 in 2025, down by a third in 15 years. Foreign workers fill the gap and now make up nearly one in four skilled shop-floor workers. Of the 11,641 people in shipbuilding on Japan's Specified Skilled Worker visa, more than half are from the Philippines.

So how many people would twice the ships take? We ran the numbers.

About 77,000 people work in Japanese shipbuilding today. Working the way they do now, twice the ships would take about 137,000. That's about 60,000 people short. Even with the government's assumed 25% productivity gain from robots and AI, it would take about 110,000.

Doubling the workforce isn't realistic. The only path is to multiply what each person can build with robots and AI, which collides head-on with Part 01: ships are exactly what robots find hardest.

One more surprise. Building the same ship costs about 20% less in China. But almost all of that gap is steel and materials; labor costs are actually slightly higher in China. Robots don't fix Japan's price problem. They fix its volume problem: how much a small workforce can build.

So how far have Japanese shipyards come? We sorted the 72 cases from our list by the job each one targets and how far along it is: in use on the shop floor, in trial, planned, or still in research.

Only 19 of the 72 are running on the shop floor, and most of them weld flat parts or are small and portable.

• Oshima Shipbuilding added five portable welding robots in 2024, for 18 in total, on its small-parts line.
• Shin Kurushima Dockyard brought in a 15kg welding robot made by System LSI, an image-inspection maker in nearby Matsuyama. It finds weld lines with a camera and laser, and one person is expected to run three.
• Onomichi Dockyard showed its first-ever industrial robot in November 2025, which finds weld lines with sensors.

Work inside cramped hull bottoms, high up, or on curved plates is all still at the research stage.

Lined up by year, the 72 cases show another trend. In 2022-2024 most were about data and digital tools such as schedules and 3D design. Welding and painting robots arrived in 2025, and in 2026 AI-driven robots took off. Of the 32 cases from 2026, 24 are research or plans. The new wave has only just begun.

AI starts learning a master's feel

JMU is furthest ahead.

The curved plates at a ship's bow and stern are shaped by a craft called line heating. A worker heats a flat steel plate along a line with a gas torch and immediately cools it with water; the heated strip shrinks and the plate bends a little. Where and in what order to heat has always depended on a master's experience and feel.

Since 2024, AI robots have been doing line heating at JMU's Tsu yard. Simulations recreate how masters heat plates, an AI trained on that data decides where to heat, and small robots bend the steel. The error is within 1cm, and one less experienced worker could do the work of three.

The robots are slower than people, and JMU says it won't automate everything. Veterans still handle the plates that need the most precision. The robots fill the gap left by too few craftsmen rather than replacing them.

Wall-climbing quadrupeds, and Toyota inside the hull

In 2026 things got bolder. The government put ¥15 billion into "AI-driven next-generation shipbuilding robots" and in March selected 14 R&D projects. For anyone who likes robots, the list is a treat.

• A quadruped that climbs steel walls. In October Kawasaki Heavy Industries showed a prototype four-legged robot that grips the hull with electromagnets and welds as it climbs. At its Sakaide yard it is working with NVIDIA on a "digital shipyard", a virtual copy of the whole yard used to generate robot motions.
• Robots that walk inside hull bottoms. MTI (part of the NYK group), JMU, Namura and Shin Kurushima are developing legged robots that weld and paint in cramped double bottoms.
• Toyota polishing ship interiors. Shin Kurushima, Imabari and Toyota are developing robots that move through compartments on their own, cleaning and polishing.
• Robots that work as a swarm. Yaskawa and FANUC are part of a project to coordinate multiple robots with AI.

The government has also teamed up with the US to build a large hull mock-up at the University of Michigan for developing and testing robots. HII, America's largest military shipbuilder, and AI welding firm Path Robotics are involved, and 12 US AI and robotics startups have expressed interest.

The competition, though, is further ahead.

Korea is competing on robots per worker. At HD Hyundai's yard in Ulsan, each worker used to run about two robots. A trial is now running with six, and the goal is eight. The robots work out where to weld from drawing data and ride self-driving carts to the next spot. At rival Hanwha Ocean, AI already supports 67% of indoor welding, with a goal of fully automated welding by 2030. HD Hyundai has also begun developing a welding humanoid with US firm Persona AI.

China is pushing volume and automation at once. At private shipbuilder Yangzijiang, welding stations that needed 8-12 people now run with about four, and welding efficiency is up 40%. A Hong Kong Polytechnic University professor told the Financial Times that Korea and Japan are about five years ahead of China on robots, but China could catch up.

The US pays for results. HII, the largest US military shipbuilder, will give AI welding firms Path Robotics and GrayMatter Robotics up to $900 million of work over seven years, released only if cost, schedule and quality milestones are met. The US Navy committed up to $448 million to Palantir's shipbuilding software, and Palantir said a pilot cut submarine schedule planning from 160 hours to under 10 minutes.

Japan only began R&D on AI shipbuilding robots in 2026. Even if a five-year lead is real, it won't last long.

So who will bring robots and AI into the yards? We read 28 shipyard tech job postings open on October 17, 2026.

25 of the 28 did not require shipbuilding experience. The three that did were for systems on the ships themselves, such as autonomous navigation.

The language is blunt. Imabari's careers page says most of its mid-career hires have never worked in shipbuilding. Kawasaki Heavy Industries says shipbuilding experience is not required and that experience at car or industrial machinery makers is welcome, citing a 31-year-old former car designer earning ¥7.5 million a year. Mitsubishi Shipbuilding is building a new hull-block factory in Shimonoseki for about ¥100 billion, and one of its postings there calls welding automation urgent.

The posting that stood out most was at JMU's Tsu yard: developing humanoid robots that work alongside people in the shipyard, to carry on master welders' and painters' skills, at ¥6-9 million a year.

Still, only a handful of the 28 are for running robots or AI on the shop floor; most are internal IT and digital roles. For what AI-in-the-field engineering jobs look like more broadly, see our reading of 326 forward deployed engineer postings.

EDITOR'S TAKE ── Kaito Ogasawara's view

The race is robots per worker

Can Japan double its shipbuilding? Lining up 72 projects and the moves abroad, the answer I see is not more people but more output from each pair of hands.

HD Hyundai is aiming for eight robots per worker. In Japan, Shin Kurushima plans one person for three robots, and JMU's line-heating robots could let one worker do the job of three. There is still a clear gap.

The aim is the same, though: not to replace craftsmen with robots, but to raise how many robots each craftsman can run. That multiplier is the only way to fill a 60,000-worker gap.

Half of automatic welding work was people teaching robots where to weld. With a different hull every time, that is the biggest wall.

Korean robots that read drawing data to find weld lines, and Path Robotics building AI just for welding, are both aimed at that wall. Whether Japanese yards can turn their 3D design data into something robots can read directly matters more, I think, than how many robots they buy.

The money is moving, and Japan has teamed up with the US. Shipyards are seriously hunting for engineers from outside the industry. On the other hand, most 2026 AI robot projects are one-year research efforts, and equipment can't be added overnight: the Shipbuilders' Association of Japan says large cranes ordered today won't arrive for more than seven years.

Even so, if engineers from cars, robotics and AI flow into shipbuilding, I think Japan's yards can change. Ships may stay handmade, but each pair of hands can do many times more. DATA WORLD will keep following shipyards and their robots, on the ground.