Embodied AI: Japan’s moment?

Robotics is approaching its PC moment. A shift from narrowly programmed machines to more general-purpose systems capable of operating in the real world is now underway. As with personal computers, the importance of this transition lies less in any single device than in the platform it creates: a new layer of capability that allows intelligence to move beyond tightly controlled environments and into everyday economic activity.

For decades, industrial robots have welded car bodies, stacked pallets, and repeated precise motions with tireless accuracy across manufacturing. Durability, speed, and precision improved steadily, but cognition did not. Traditional robots executed instructions without reason, judgement, or autonomy. That constraint is now collapsing. With modern artificial intelligence, the brains have finally caught up with the bodies. AI is becoming embodied.

As AI pushes into the physical world, attention has centred on so-called humanoid robots. That focus attracts scepticism, and not without reason. Debates about form factor, however, risk obscuring the substance of what is changing. The real shift is not the shape of the machine, but the arrival of intelligence that can perceive, reason, and act in the physical world.

“The real shift is not the shape of the machine, but the arrival of intelligence that can perceive, reason, and act in the physical world”

Whether these systems walk on two legs, roll on wheels, or take other forms entirely is secondary. What matters is that embodied systems can operate autonomously outside tightly scripted environments and integrate directly into factories, warehouses, and other real-world settings where economic activity actually takes place.

Lessons from past platform transitions

History is unkind to those who miss platform transitions. In the PC era, the largest rents accrued not to hardware champions but to those who controlled operating systems and ecosystems. Microsoft mattered more than Intel. In smartphones, Apple eclipsed Nokia by owning the integration of hardware, software, and services. The same logic is likely to govern embodied AI. Hardware will proliferate, but software, data, and services will determine the winners.

Industrial robots today resemble the mainframes of an earlier era: powerful but narrowly bounded. Embodied AI represents the shift to general-purpose systems. As before, those who cling to hardware excellence without controlling the software and learning layer risk being left behind.

Corporate and National imperatives

The race has already begun. In the United States, embodied-AI efforts are tightly coupled to frontier software and AI platforms. In China, the strategy is speed and scale, with ambitions to deploy tens of millions of units over the next few years. If realised, that scale would reset global cost curves and place immediate pressure on manufacturers everywhere to adopt similar technologies.

China’s push matters because it starts the clock. In our view, this will not be a discretionary transition. It will be forced by economics. As unit costs fall and capabilities improve, manufacturers will face a simple choice: adopt or fall behind.

That urgency exposes a deeper problem. Autonomous robots, whatever their shape, make perfect spies. They observe, measure, and quantify everything they interact with. Every adjustment, workaround, and optimisation can be captured, transmitted, and analysed far from the factory floor. What begins as productivity enhancement can quickly become systematic extraction of operational know-how.

This is how platform transitions concentrate power: through aggregating learning. If embodied AI becomes a general-purpose industrial platform, leadership in manufacturing may increasingly depend on who controls the software systems trained on that learning. Someone will own that layer.

For individual companies, the imperative is strategic. They cannot afford to hand their most valuable asset, operational know-how, to an external software owner in exchange for short-term gains. For manufacturing nations, the imperative is more fundamental. Countries whose prosperity rests on industrial excellence cannot afford to see that knowledge centralised and monetised elsewhere.

In the near term, China’s scale push compresses costs and forces adoption. Over the longer term, the risk shifts: control migrates to the software layer. Own it, or steadily lose manufacturing advantage.

Japan’s hidden advantage

Japan’s depth across the physical robotics stack is widely recognised as extraordinary. In machine tools, the so-called mother machines of industrial automation, Japanese firms sit at the heart of global production capacity. In motion control, precision motors, harmonic drives, and ultra-high-precision gearing, Japan again leads globally. Beneath that layer sit world-class bearing manufacturers and industrial component specialists that quietly underpin high-precision machinery worldwide.

The stack extends far beyond mechanics. Robots are dense electro‑mechanical systems, and Japan dominates many of the components that make them work: power semiconductors governing motor control and efficiency, multilayer ceramic capacitors stabilising power delivery, precision connectors and wire harnesses forming the nervous system of machines, and a wide array of advanced industrial sensors.

Vision and sensing are another area of strength. Embodied AI depends on perception, and Japan sits at the centre of global opto‑electronics. Sony’s leadership in advanced image sensors, alongside deep capabilities in optics, lasers, photonics, and industrial imaging systems, places Japanese firms at the core of how machines see and interpret the physical world.

Materials science adds a further layer. Advanced ceramics, specialty steels, carbon-fibre composites, and high‑performance industrial materials underpin durability, precision, and thermal stability in robotics. Japanese companies have spent decades refining these capabilities in specialised niches where process knowledge compounds over time.

Taken together, Japan does not simply participate in robotics supply chains. In many places, it anchors them.

And yet, despite being the world’s largest buyer of industrial robots and a major supplier to that same market, Japan is not as front and centre in the global embodied-AI debate as this natural endowment would suggest. There is no Japanese champion commanding attention at the platform level. Honda retired its pioneering robot, ASIMO, just as large-scale generative AI arrived.

The issue is not inactivity or capability. Rather, it is fragmentation and focus. Efforts remain spread across institutions with no single organising layer. Moreover, they remain largely weighted toward hardware rather than the software systems that aggregate learning and compound advantage.

Paradoxically, within Japan’s hardware strength lies significant latent potential on the software side. Japan operates one of the largest installed bases of industrial robots, machine tools, and precision equipment in the world. At scale, these deployed systems generate exactly the data required to train perception, planning, and decision systems for embodied AI.

“Japan operates one of the largest installed bases of industrial robots, machine tools, and precision equipment in the world”

Maintaining leadership in hardware is no longer sufficient. Japan now needs to organise around its installed base, aggregate learning across deployments, and turn that scale into a compounding software advantage.

Execution at scale

Japan is not short of software ambition. A number of Japanese startups are attacking elements of the embodied AI stack across multiple industrial verticals. These efforts are impressive and necessary, and they merit close attention.

However, the scale of activity, capital, and patience required to build an end-to-end embodied AI platform is formidable. This is unlikely to be solved by incremental startup experimentation alone, or by additional government coordination. Animal spirits must be combined with scale: the willingness to invest ahead of returns, tolerate long deployment cycles, and sustain learning systems through years of expensive iteration.

Recent experience in automotive autonomy is instructive. At the start of this decade, enormous amounts of capital flowed into ADAS (Advanced Driver Assistance Systems) and autonomous driving, driven by a mix of disruptive startups and incumbent OEMs (Original Equipment Manufacturers). Much of that effort disappointed. Many sensor and rule-driven approaches struggled to generalise beyond constrained environments, while OEM-led programmes found it difficult to reinvent themselves as builders of continuously learning software platforms.

More durable progress has come from software-native approaches built around end-to-end learning and generalisation. These systems treat autonomy as a foundation-model problem, learning directly from sensor inputs to actions and improving through deployment rather than hand-coded updates. Competitive advantage emerges from learning velocity and data efficiency as systems scale.

This lesson matters directly for embodied AI. Building general-purpose physical intelligence requires software architectures that can learn from messy real-world data, adapt across environments and tasks, and compound improvement over time. That capability is slow to build, expensive to sustain, and unforgiving of partial execution. Without it, even large installed bases struggle to translate scale into durable platform advantage.

Japan’s potential leaders

Capability is necessary, but not sufficient. The question is not whether Japan has the raw ingredients to lead in embodied AI, but which institutions can organise that capability into a scalable, software-led platform. There are candidates.

SoftBank Group is one. Through its acquisition of ABB Robotics, the technology-focused conglomerate has effectively placed software-native capital and platform thinking on top of one of the world’s largest industrial-robotics installed bases. At the same time, its ownership stake in Wayve gives it exposure to one of the most coherent end-to-end learning architectures emerging in autonomy and embodied AI. Few institutions globally have positioned themselves across both large-scale physical deployment and frontier learning systems in this way. If embodied AI becomes a platform industry, SoftBank has quietly assembled several of the ingredients required to participate at that level.

Hitachi is another. Through Lumada, it has built one of the world’s largest industrial AI and data platforms, spanning data integration, analytics, digital twins, and domain-specific applications across rail, energy, factories, and infrastructure. Lumada is not an experiment; it is a substantial business of roughly ¥4 trillion in revenue with deep enterprise penetration, demonstrating that Japanese industry can build and operate software platforms at scale.

What neither SoftBank nor Hitachi yet has, however, is sustained exposure to the kind of high-volume, safety-critical, real-time physical environments where embodied AI systems learn fastest. That exposure is not optional. It is where learning velocity is forged and software architectures are tested against reality rather than controlled use cases.

Toyota: A call to arms

In that context, Toyota Motor Corporation stands apart. Toyota has already spent much of the past decade grappling with the hardest version of this problem through ADAS and autonomous driving. That journey has been slow, expensive, and at times disappointing. While Toyota has not yet solved self-learning autonomy, it is deep into the rabbit hole, accumulating organisational muscle memory and building a growing network of external partnerships.

“Toyota…could become the organising force for Japanese robotics and embodied AI, mobilising a broad swathe of the country’s hardware, software, and systems expertise.”

Crucially, Toyota brings something no other Japanese institution can match: the ability to industrialise. Its manufacturing scale, supplier network, logistics infrastructure, quality systems, and global deployment capability allow it to replicate complex technologies reliably across millions of units, factories, and environments.

If Toyota can extend this industrial model from autos into autonomous robotic systems and layer a learning software platform on top, it could become the organising force for Japanese robotics and embodied AI, mobilising a broad swathe of the country’s hardware, software, and systems expertise.

Should Toyota take this on and succeed, the implications would extend well beyond a single corporate winner. It would reinforce Japan’s manufacturing base, deepen competitive moats across the supply chain, and deliver meaningful second-order benefits for the Japanese economy and equity market.

Is this wishful thinking? Perhaps not. Through its Research Institute, Toyota is already collaborating with global robotics firm Boston Dynamics on a research basis, providing its large behaviour model (LBM) to the Atlas humanoid. Toyota already holds substantive IP in embodied AI and is developing it directly in a robotic context.

It is reasonable to ask what might emerge if Toyota’s industrial scale were paired with the software-native capital, platforms, and technological capabilities of companies such as SoftBank, Hitachi, Sony, Fanuc and others across Japan’s broader industrial ecosystem.

Conclusion: Will Japan seize the moment?

AI is opening a vast opportunity in robotics. This is not an incremental extension of factory automation but a step-change in what machines can do in the physical world. Japan is well positioned for this moment. It has the industrial know-how, component depth, manufacturing discipline, and supply chains that matter. Few countries begin with such an advantage.

However, Japan is not yet moving with the same speed or cohesion as the United States or China. That is the strategic choice now facing Japan. If it drifts, others may define the platforms, standards, and learning loops that shape the next era of manufacturing. We expect Japan to mobilise. Ideally, Toyota could emerge as the tip of the spear, not as a lone champion but as the organising force capable of turning Japan’s deep industrial and robotics strength into a global embodied-AI platform.

The scale of the prize should be motivation enough. The cost of failing to act, once the window has closed, would be far greater.