Japanese companies intensify “Physical AI”: from single-model competition to manufacturing scenario competition

A new signal in Japan’s AI industry is emerging at the capital level.

According to Kyodo News, Fujitsu, Asahi Kasei, and Yaskawa Electric are considering investing in a new company led by SoftBank, Japan AI Foundation Model Development. The report says the three companies are expected to invest tens of millions of yen each, and will work with the new company to advance a direction known as “physical AI”: combining AI with robotics and applying it in manufacturing scenarios to make robots on production lines more autonomous.

On the surface, this is an expression of corporate investment intent. In essence, however, it reflects a structural shift in Japan’s AI strategy. Japan does not intend to confront the United States and China head-on solely by competing in the training scale of general-purpose large models. Instead, it is trying to leverage its accumulated manufacturing process data, equipment ecosystem, and robotics industry base to carve out a more differentiated AI path in industrial settings.

This is also a pragmatic choice for Japan’s technology industry.

In the global AI race, the United States and China have established clear leading advantages in model scale, compute investment, and platform ecosystems. For Japan, simply copying this route would be both costly and unlikely to create a comparative advantage. By contrast, Japan has long-standing strengths in manufacturing, inspection, precision machining, industrial automation, and robotics integration. These capabilities are not easily replaced by general internet companies in the short term. Embedding AI into factories, equipment, and production processes can turn Japan’s traditional industrial assets into scenario advantages for AI training and deployment.

Kyodo’s report noted that the new company hopes to use large volumes of data accumulated in Japanese manufacturing sites, focusing on training models related to inspection and processing. This is crucial. For manufacturing AI, what determines system capability is not just the algorithm itself, but whether the data comes from real industrial processes and covers exception handling, quality control, process changes, and equipment coordination. If such data can be used in a structured way, Japan’s manufacturing sector may have the opportunity to transform “tacit experience” into reproducible intelligent capabilities.

From the perspective of industrial division of labor, the mix of participating companies is also representative. SoftBank, NEC, Honda, and Sony are the main shareholders, while institutions such as MUFG Bank and Nippon Steel are also involved, and Fujitsu, Asahi Kasei, and Yaskawa Electric are considering joining. This lineup shows that Japanese companies do not view AI as a standalone IT project, but rather understand it within a broader industrial-chain framework: financial capital, industrial manufacturing, materials, robotics, automotive, electronics, and infrastructure companies all participating together, seeking to build a cross-industry AI alliance.

What is most worth noting is the term “physical AI” itself.

It means that Japanese companies’ understanding of AI is shifting from “generating content” to “controlling actions.”What is most worth paying attention to is the very phrase “physical AI.”

It means Japanese companies’ understanding of AI is shifting from “generating content” to “controlling actions.” In consumer internet scenarios, the value of large models lies mainly in text generation, knowledge Q&A, and office automation; but in manufacturing scenarios, what truly determines competitiveness is whether AI can enable robots to identify defects, adjust movements, adapt to changes in the environment, and maintain stability in complex processes. In other words, AI is no longer just a cognitive tool; it is becoming part of industrial control systems.

This is especially important for industrial robot companies like Yaskawa Electric. Industrial robots have long been widely used in welding, handling, assembly, and inspection, but traditional automation systems still have limited flexibility when faced with manufacturing environments involving many product types, small batches, and frequent changeovers. If AI can improve robots’ ability to perceive environmental changes and make decisions, Japan’s manufacturing automation upgrade could move from “fixed-program automation” to “adaptive automation.”

For Japan’s manufacturing industry, the significance of this shift goes beyond efficiency gains.

Japan has long faced problems such as a declining working-age population, shortages of skilled workers, and difficulties in factory succession. If the combination of AI and robots can share labor burdens in inspection, processing, and on-site judgment, it may become an important tool for easing labor constraints. More importantly, such applications do not rely on abstract breakthroughs in “general intelligence”; instead, they can be gradually implemented in concrete industrial scenarios, creating verifiable commercial value.

The new company’s application to Japan’s New Energy and Industrial Technology Development Organization for funding support also shows that this is not an isolated corporate move, but one that resonates with the direction of national industrial policy. In recent years, the Japanese government has been emphasizing the rebuilding of semiconductors, digitalization, automation, and advanced manufacturing, and AI is gradually becoming the connective layer for these strategies: it is both a software capability and a tool for manufacturing upgrades, as well as infrastructure for industrial competitiveness.

From an investment perspective, the willingness of several large Japanese companies to take minority stakes with relatively limited amounts reflects that their attitude toward AI has shifted from waiting and watching to collaborative construction. Compared with developing systems independently, a more feasible path may be to share models, share data standards, and share scenario validation, then form division of labor through practical applications in each industry. This approach aligns with the alliance-based innovation that Japanese companies have traditionally preferred, and may also be better suited to the current stage than a single company’s isolated breakthrough.

However, this route also faces real challenges.

First, although manufacturing data is abundant, it is often scattered across different factories, different equipment, and different enterprise systems, making standardization difficult. Second, the barrier to deployment for industrial AI lies not only in model capability, but also in deep integration with production line control, robotic systems, safety standards, and on-site maintenance systems. Finally, even if “physical AI” first develops competitiveness in Japan’s domestic market, how to further expand into overseas manufacturing systems will still determine its long-term commercial potential.

But in any case, this trend has already shown that Japan’s AI competition does not intend to simply repeat the Silicon Valley path.

Japan is more likely to choose a route highly matched to its own industrial structure: using manufacturing as the scenario, robots as the execution end, industrial data as the training foundation, and enterprise alliances plus policy support as the organizational method.Japan is more likely to choose a path that closely matches its own industrial structure: using manufacturing as the setting, robots as the execution end, industrial data as the training foundation, and corporate alliances plus policy support as the organizing mechanism. This is not a rejection of the trend toward general-purpose large models, but rather a shift in the center of AI competition from “who can train the bigger model” to “who can turn AI into real productive force.”

If this path can continue to advance, Japan’s technology industry may not stand out in the most visible model rankings in the global AI landscape, but it may gradually emerge in factories, production lines, and industrial equipment. This is exactly Japan’s most familiar—and most adept—mode of competition: not making a splash first, but embedding technology deep into the industrial chain and waiting for it to be transformed into long-term advantage.