Physical AI and the Japanese Market: Silicon, Software, Factories, and Robots

The third wave of AI has a body

The first wave of modern AI lived in data. The second — generative AI — lived in language and images. The third, Physical AI, lives in the world: intelligence embodied in machines that perceive their surroundings, reason about them, and *act* — humanoid robots, autonomous systems, and AI-driven factories that manipulate physical reality rather than just text on a screen.

No country has more reason to lead this wave than Japan — and few are moving as deliberately. For most economies, physical AI is an efficiency lever. For Japan, it is closer to an existential necessity.

---

Why Japan, why now: demographics as destiny

Japan's population has declined for 14 consecutive years as of 2024. By 2030, roughly 30% of the population will be over 65. Around 1.2 million jobs already sit unfilled, and the Recruit Works Institute projects a labor shortage of 11 million people by 2040. In elder care alone, Japan faces a shortfall of 570,000 care workers by 2040.

This reframes the whole conversation. As one report put it, the robot in Japan "isn't coming for your job; it's filling the one nobody wants." Robots are already lifting patients, delivering meals, and monitoring vital signs overnight so nurses can focus on care — and humanoids have worked as baggage handlers at Japanese airports. Necessity, not novelty, is the driver.

---

The market: small today, steep tomorrow

The numbers reflect an inflection point:

Global physical AI: ~$5.0 billion in 2025 → ~$82.8 billion by 2034 (CAGR ~32.8%).
Global humanoid / embodied AI robotics: $3.77 billion (2025) → $34.37 billion (2033) at ~29.5% CAGR.
Japan's humanoid robot market specifically: $0.22 billion (2025) → $3.99 billion (2034) — a striking 43.7% CAGR, among the fastest of any national segment.
Momentum check: global humanoid shipments are set to exceed 50,000 units in 2026, up ~178% from roughly 12,000 in 2025.

---

A national strategy, not a market bet

In December 2025, Japan's Cabinet approved the Basic National AI Plan — the first national framework to formally designate Physical AI and humanoid robotics as sovereign strategic priorities. The plan is structured to catalyze over ¥50 trillion (~$327 billion) in combined public-private investment by 2030, with allocations reported around ¥3.9 trillion for domestic semiconductor and AI-chip production and ¥1.2 trillion for robotics R&D, automation, and edge-AI deployment. Japan's stated ambition: capture 30% of the global physical AI market by 2040.

Physical AI needs a full stack to work — silicon to compute, embedded software to control, bodies to act, and the intelligence to bind them. Japan is deliberately building every layer.

---

Layer 1 — The bodies: robots and humanoids

Japan starts from strength. It holds the world's highest robot density, and its "Big Five" industrial robot makers — Fanuc, Yaskawa, Kawasaki Heavy Industries, DENSO, and Mitsubishi Electric — together account for over 40% of global industrial-robot shipments.

That base is now pivoting toward embodied, general-purpose machines:

Kawasaki Kaleido 9 (unveiled late 2025): a 180 cm, 86 kg humanoid with 41 degrees of freedom that can lift 25 kg per arm — among the most capable Japanese-designed humanoids yet, aimed at assembly and logistics.
The KyoHA consortium (launched July 2025) pools Waseda University, tmsuk, Murata, Renesas, and Sumitomo Heavy Industries, each contributing a piece — sensors, motors, control MCUs, AI — to build an *integrated domestic humanoid supply chain*.
Preferred Networks and Toyota are jointly developing service robots, pairing frontier AI with Toyota's manufacturing and mobility scale.
A further consortium of ~20 manufacturers is set to form in October 2026 to develop AI that lets machine-tool and factory robots perform tasks with far more autonomy.

Leading Japanese humanoid players also include SoftBank Robotics, Honda, Panasonic, Mitsubishi Heavy Industries, Kawada Robotics, Cyberdyne, and Toshiba.

---

Layer 2 — Where physical AI lands first: the smart factory

Humanoids grab headlines, but the near-term center of gravity is the smart factory — the environment where physical AI is most immediately deployable and profitable.

Japan's smart-factory market is projected to reach ~$9.2 billion by 2034 (CAGR ~9.0% over 2026–2034), while its digital-twin manufacturing market is set to hit ~$16.1 billion by 2033. The playbook is consistent: AI-driven production, digital twins, predictive maintenance, collaborative robots, and smart sensors optimizing processes in real time.

Concrete moves are already on the ground. Kyocera committed ¥66 billion to a 150,000 m² smart factory in Nagasaki (started September 2024) producing semiconductor packages and fine-ceramic components. And vendors like Ryoyo Electro are building digital-twin platforms with partners including NVIDIA and avatarin — the same NVIDIA-plus-Japanese-industrial pattern (Fujitsu, Yaskawa) recurring across the sector. The smart factory is, in effect, physical AI's proving ground.

---

Layer 3 — The silicon foundation: chip manufacturing

Physical AI is hungry for edge silicon — the chips that let a robot see, decide, and move in milliseconds without a round trip to the cloud. That is why Japan's semiconductor renaissance is inseparable from its physical-AI strategy. The government has committed over ¥10 trillion in public support for semiconductors and AI by FY2030.

Two projects anchor the effort:

Rapidus — Japan's national champion for leading-edge logic — secured $1.7 billion in fresh funding (including ¥100 billion from the IPA and ¥167.6 billion from 32 private firms such as Canon, Fujitsu, NTT, SoftBank, and Sony) to chase 2nm mass production by 2027. Its 2nm pilot line went live in April 2025, using the world's most advanced High-NA EUV lithography. Estimated government backing approaches ¥2.9 trillion.
TSMC's JASM in Kumamoto — a venture with Sony, Denso, and Toyota as shareholders — has total investment exceeding $20 billion, with a second fab advancing toward more advanced nodes (reported 3nm) and mass production expected around 2028; Japan approved subsidies of up to $4.62 billion for that second facility, with ~¥1.2 trillion tied to Kumamoto overall.

Notably, the shareholder lists read like a physical-AI roster — Toyota, Denso, Sony appear on both the robot and the chip side. Japan is vertically integrating the entire stack, from wafer to humanoid.

---

Layer 4 — The connective tissue: embedded and real-time software

A chip is inert and a robot is dumb without the code between them: embedded, real-time software that reads sensors, enforces safety, and drives actuators on deterministic deadlines measured in microseconds. This is a domain where Japan has quietly deep roots, forged over decades in automotive and industrial control.

eSOL, a 51-year-old Tokyo firm, builds real-time operating systems (RTOS), middleware, and tooling for automotive, industrial, medical, and consumer systems — and is an active contributor to both the AUTOSAR standard and the Autoware Foundation's open-source autonomous-driving stack. Renesas — already a KyoHA member *and* a Rapidus backer — is pushing automotive SoCs for ADAS and the software-defined vehicle, spanning RTOS, QNX, Linux, and Android.

The discipline these teams live by — determinism, functional safety (ISO 26262), and long-lifecycle stability — is exactly what embodied physical AI demands. A humanoid on a factory floor, or a robot lifting a patient, is a safety-critical real-time system first and an "AI product" second. Japan's automotive-grade embedded culture is a genuine, if under-celebrated, advantage.

The catch is talent. The embedded skills gap is global and acute: fluency in C/C++, RTOS internals, and standards like AUTOSAR and ISO 26262 is scarce everywhere — and Japan's shrinking workforce sharpens the squeeze. The same demographic force driving demand for physical AI is thinning the ranks of the engineers who must build its nervous system.

---

The honest challenges

The strategy is coherent, but not guaranteed. Rapidus must still scale from pilot to volume, secure sustained private investment, and — hardest of all — win customers against an entrenched TSMC. JASM has weathered reported losses on its way up the node ladder. Humanoids remain expensive and are only beginning the leap "from pilot to platform." And the same demographic crunch driving demand also starves the talent pipeline needed to build these systems. Physical AI in Japan is a bet placed under real pressure — which is precisely what makes it credible.

---

The outlook

Japan is doing something unusual: treating physical AI not as a single product category but as a full-stack national project — silicon at the bottom, smart factories in the middle, robots and humanoids at the top, all pulled forward by a demographic reality no policy can reverse. If any country can turn "the world's oldest society" into "the world's most automated," the pieces are now visibly on the board.

---

日本語まとめ：フィジカルAIと日本市場

AIの「第3の波」であるフィジカルAI（Physical AI）は、データやテキストの中ではなく、現実世界で知覚・判断・行動する知能です。ヒューマノイド、自律システム、そしてAIが物理世界を動かす工場——これがその舞台です。そして、これを牽引する必然性を最も強く持つ国が日本です。多くの国にとって効率化の手段であるフィジカルAIは、日本にとっては存続に関わる必要性に近いものです。

なぜ日本か：人口動態

日本の人口は2024年時点で14年連続の減少。2030年には約3割が65歳以上になります。すでに約120万件の求人が埋まらず、リクルートワークス研究所は2040年に1,100万人の労働力不足を予測。介護分野だけでも2040年に57万人の担い手が不足します。ロボットは「仕事を奪う」のではなく「誰もやりたがらない仕事を埋める」存在になりつつあります。

市場規模

世界のフィジカルAI市場：2025年約50億ドル → 2034年約828億ドル（CAGR 約32.8%）
日本のヒューマノイド市場：2025年 2.2億ドル → 2034年 39.9億ドル（CAGR 43.7%）
世界のヒューマノイド出荷：2026年に5万台超（前年比約+178%）

国家戦略

2025年12月、政府は「AI基本計画」を閣議決定し、フィジカルAIとヒューマノイドを国家戦略の重点と初めて明記。2030年までに官民50兆円超の投資を狙い、半導体・AIチップに約3.9兆円、ロボR&D・エッジAIに約1.2兆円を配分。2040年に世界市場の30%獲得を目標に掲げます。

フィジカルAIに必要なのは、シリコン（計算）・組込みソフト（制御）・身体（動作）・知能（統合）というフルスタック。日本はこの全層を同時に整備しています。

各層

身体（ロボット）：世界一のロボット密度。ファナック・安川・川崎重工・デンソー・三菱電機の「ビッグ5」が世界の産業用ロボット出荷の4割超。川崎重工のKaleido 9（2025年末、41自由度・片腕25kg）、産学連携のKyoHAコンソーシアム（早稲田・tmsuk・村田・ルネサス・住友重機）、Preferred Networks × トヨタのサービスロボット、2026年10月発足予定の約20社によるAIコンソーシアムなど。
工場（スマートファクトリー）：日本のスマートファクトリー市場は2034年に約92億ドル、製造業のデジタルツインは2033年に約161億ドルへ。京セラは長崎に660億円の先端工場を建設。NVIDIAと日本の産業各社（富士通・安川・菱洋エレクトロ等）の連携が加速。
シリコン（半導体）：エッジ推論を支えるチップが鍵。政府は2030年度までに半導体・AIへ10兆円超を投じる方針。ラピダスは17億ドルを調達し2027年の2nm量産を目指す（2025年4月に2nmパイロットライン稼働、High-NA EUV採用）。TSMC熊本（JASM）はソニー・デンソー・トヨタが出資し、総投資200億ドル超。トヨタ・デンソー・ソニーがロボットと半導体の両側に登場する点が象徴的です。
組込みソフト（つなぐ層）：チップとロボットの間で、センサーを読み、安全を担保し、マイクロ秒単位でアクチュエータを制御するリアルタイム組込みソフトがなければ、シリコンもロボットも動きません。日本はここに長い蓄積があります。eSOL（創業51年、RTOS・AUTOSAR・Autoware Foundation）、ルネサス（車載SoC・ADAS・ソフトウェア定義車両）など。決定論・機能安全（ISO 26262）・長期安定性という車載由来の規律は、身体を持つフィジカルAIにそのまま効きます。課題は人材——C/C++やRTOS、AUTOSAR/ISO 26262に精通した技術者は世界的に不足し、人口減の日本では一層深刻です。

課題

ラピダスの量産化と顧客獲得、ヒューマノイドの高コスト、そして需要を生むのと同じ人口減が人材の供給を細らせるというジレンマ。それでも、圧力の下で置かれたこの賭けだからこそ現実味があります。

日本は、フィジカルAIを単一製品ではなく、シリコン→工場→ロボットという全スタックの国家プロジェクトとして進めています。「世界で最も高齢化した社会」を「世界で最も自動化された社会」へ——その駒は、いま盤上に並び始めました。

---