▶️ LISTEN TO THIS ARTICLE

Japan will be short 11 million workers by 2040. Not hypothetically. The country's working-age population has already fallen 16% from its 1995 peak of 87.3 million to 73.7 million in 2024, and the decline is accelerating. In December 2025, Prime Minister Sanae Takaichi's cabinet approved Japan's first-ever national AI plan, openly acknowledging what Tokyo had been reluctant to admit: Japan has fallen behind in AI investment, commercialization, and talent. The fix they're proposing costs $19 billion in combined public-private spending, including a new national AI company tasked with building a trillion-parameter foundation model from scratch.

Whether Japan can turn robotics expertise and capital deployment into actual AI relevance is the question the money is supposed to answer.

SoftBank's Everything Bet

No company embodies Japan's AI ambitions more than SoftBank Group. Masayoshi Son's firm has become the single largest private investor in the AI supply chain, and the numbers are staggering. SoftBank completed a $41 billion investment in OpenAI in late 2025, securing roughly 11% of the ChatGPT maker. It co-founded the Stargate Project, a $500 billion joint venture with OpenAI, Oracle, and Abu Dhabi's MGX to build AI data center infrastructure across the US. SoftBank and OpenAI each committed $19 billion in initial capital and hold 40% ownership stakes.

Then came the robotics play. In October 2025, SoftBank agreed to acquire ABB's robotics division for $5.4 billion, a unit with 7,000 employees and $2.28 billion in 2024 revenue. Son described the move as the start of "Physical AI," his term for merging robotics with what he calls artificial superintelligence. In January 2026, OpenAI and SoftBank invested $1 billion in SB Energy to build multi-gigawatt data center campuses.

The strategy is clear if aggressive: SoftBank wants to sit at the intersection of AI compute, AI software, and AI embodiment. It doesn't build foundation models itself. It buys stakes in the companies that do, then tries to integrate those models into physical systems.

Physical AI: Where Japan's Real Advantage Lives

"Physical AI" was the dominant theme at iREX 2025, Japan's flagship robotics exhibition, and for good reason. Japan's industrial robotics companies are doing something their American and Chinese competitors can't easily replicate: teaching machines to see, feel, and adapt.

Japan needs AI and robotics to function as a society in 15 years. That urgency gives its investments a different character.

Fanuc partnered with NVIDIA to develop factory robots that respond to spoken commands and use visual generative AI to perceive depth and occlusion like a human operator. If a part slips during handling, the robot senses the shift through visual and tactile feedback and adjusts its grip in real time. These aren't pre-programmed motions. They're learned behaviors.

Yaskawa Electric and SoftBank began collaborating on Physical AI to bring autonomous robots into office environments, not just factory floors. Yaskawa's 2026 systems learn through demonstration and haptic feedback, using reinforcement learning to master tasks that were previously impossible to automate because they required human judgment about pressure, position, and timing.

The concept extends beyond manufacturing. Japanese hotels already deploy AI robots at reception desks because they can't hire front desk staff. AI-enabled delivery robots handle food courier routes. Japan is deploying AI robots in shipyards by 2026 to counter labor shortages in an industry where the average worker age keeps climbing. The technology "digitizes" the intuition of skilled workers, preserving a master welder's technique in a neural network after the welder retires.

Building a National Foundation Model

Japan's national AI plan isn't just about robots. The government committed 1 trillion yen ($6.34 billion) over five years starting in fiscal 2026 to support a new public-private AI company. The entity will employ roughly 100 engineers, primarily from SoftBank and AI startup Preferred Networks, and its first target is a 1-trillion-parameter model comparable to leading global systems.

SoftBank plans to invest a separate 2 trillion yen ($12.7 billion) in data centers for the project over six years. Meanwhile, Japan's existing players are already shipping. NTT launched tsuzumi 2 in October 2025, a lightweight LLM that runs on a single GPU while matching larger models on Japanese-language tasks, directly addressing the cost and energy concerns that make massive models impractical for most enterprises. Fujitsu's Takane LLM is being piloted in government agencies to automate policy analysis, with broader availability planned for fiscal 2026. Preferred Networks built PLaMo, a Japanese-language foundation model trained entirely from scratch.

The domestic model push reflects a growing unease with dependence on American AI. When your entire economy runs through OpenAI's API, you're one policy change or pricing decision away from disruption. Japan wants alternatives, the same sovereign AI instinct driving China's massive state-led investment and the year's broader agent-building wave.

The Innovation-First Regulatory Gamble

Japan's AI Promotion Act, approved May 2025 and effective June 4, takes the opposite approach from Europe. Where the EU AI Act creates detailed compliance obligations sorted by risk tier, Japan's law imposes exactly one obligation on the private sector: cooperate with government-led AI initiatives. There are no explicit penalties.

A model that can't match GPT-5 on benchmarks but can run an autonomous factory line has a different kind of value.

The law establishes an AI Strategic Headquarters chaired by the Prime Minister for cross-ministerial coordination and relies on existing sector-specific regulations rather than creating new AI-specific rules. As CSIS analysis noted, the framework reflects Japan's goal of becoming the "most AI-friendly country" in the world.

The bet is straightforward. Japan's private AI investment has lagged far behind the US and China. Minimal regulation is meant to change that by reducing barriers to experimentation and deployment. Whether light-touch governance holds up as AI systems handle healthcare decisions, autonomous vehicle routing, and financial analysis is a different question, one Japan has explicitly chosen to answer later rather than now. It's the opposite end of the spectrum from the EU AI Act's prescriptive framework that German and French companies are already pushing back against.

Demographics as Destiny

Every AI decision Japan makes exists in the shadow of a population crisis with no modern precedent. The country's working-age population will drop below 60 million by 2040. The old-age dependency ratio is projected to reach 74% by 2060. Seven essential service categories, from transportation to healthcare to construction, won't have enough workers to maintain current service levels. South Korea faces a parallel demographic cliff, with the world's lowest fertility rate driving similar urgency around AI-powered automation.

Research from Accenture estimates AI could inject 197 trillion yen ($1.3 trillion) into Japan's economy by 2030, and that Japan stands to gain more from AI-driven growth than any other developed economy precisely because its labor shortage is so severe. The Ministry of Internal Affairs projects Japan will need 4.98 million workers in AI and robotics by 2040 but can supply only 1.72 million at current training rates, leaving a gap of 3.26 million.

This isn't an optimization problem. It's an existential one. Japan needs AI and robotics to function as a society in 15 years. That urgency gives its AI investments a different character than the US or China, where AI is about competitive advantage and market capture. For Japan, it's about keeping the lights on.

Can Money Buy Relevance?

Japan's $19 billion plan, SoftBank's sprawling investment portfolio, and decades of robotics excellence add up to a credible but uncertain path back to tech relevance. The country excels at hardware integration, precision engineering, and manufacturing automation. It struggles with software speed, startup culture, and attracting global AI talent.

The foundation model effort is the biggest test. Building a competitive trillion-parameter model with 100 engineers is ambitious to the point of implausibility when OpenAI, Google, and Anthropic each employ thousands of researchers. Japan's advantage would have to come from integration: connecting domestic models to the physical AI systems that Fanuc, Yaskawa, and SoftBank's ABB acquisition make possible. A model that can't match GPT-5 on benchmarks but can run an autonomous factory line has a different kind of value.

Japan's AI story isn't about catching up in the frontier model race. It's about whether a country with world-class robots and a disappearing workforce can wire intelligence into its industrial base fast enough to matter.

Sources

Research Papers:

Industry / Case Studies:

Commentary:

Related Swarm Signal Coverage: