Why NVIDIA Can Break Into Japan's Robotics Market While Foreign Humanoid Products Cannot

FANUC and Yaskawa Electric are two of the world's four largest industrial robot makers. Sony holds more than half the global image-sensor market, and Nabtesco and Harmonic Drive dominate the precision reducers that drive robot joints. Lined up this way, Japan looks rock-solid in the age of physical AI. But the intuition is only half right — and that half-misalignment is exactly what decides whether foreign entrants succeed.

Strong in the "body," weak in the "brain" and the "finished product"

Split physical AI into three layers: components (the body), AI foundation models (the brain), and the finished humanoid plus its on-site deployment (the integration layer). Japan is world-class in only the first. In 2025, Chinese makers took roughly 90% of global humanoid shipments — AgiBot ~5,168 units and Unitree ~4,200 — while U.S. players like Tesla, Figure, and Agility each stayed near 150 [1]. Japan barely registers among the finished-product leaders, the legacy of major firms that built humanoids decades ago and then withdrew. In short: a world-class supplier of the body, but not the holder of initiative in the brain or the finished product.

The foreign players who succeed are selling the "brain"

Line up the clear success stories and they converge on one point: they are not selling finished products — they are selling brains.

NVIDIA leads. It embedded its Jetson edge computer into FANUC's control system and integrated its Isaac Sim platform with FANUC's ROBOGUIDE, and now FANUC, ABB, Yaskawa, and KUKA — majors with a combined installed base of over two million robots — all run NVIDIA's Omniverse, Isaac, and Jetson [2]. Google follows: in May 2026 FANUC built a Gemini Enterprise–powered AI-agent system that understands natural-language instructions and coordinates multiple robots. The pattern is the same — slot the brain inside Japan's strong body.

Foreign finished humanoids, by contrast, are still at the doorstep — taking Japanese investment or appointing local distributors, but none yet operating at scale in Japan. No foreign company has conquered the market with a finished product.

Why finished products can't get in — the "local layer"

The core: a humanoid "finished product" is not actually finished. The machine is only part of the value; what remains is the local craftsmanship of installing it, guaranteeing its safety, repairing it, and seeing the investment to payback. In Japan, that local layer is a high wall at every turn.

There is a shortage of system integrators (SIers) to install and tune robots on floors where every facility differs. There is a vacuum in safety standards — so stark that a specialist firm, Play Robotics, was founded in April 2026 specifically to fill it, with certifications like ISO 13482 and ISO/TS 15066 now gating selection [3]. And Japanese buyers demand exceptional reliability, multi-year parts guarantees, and corporate backing — bars a foreign startup that might vanish in a few years cannot clear.

Value lives in the scarcest layer

This is why NVIDIA and Google won: they did not shoulder the local layer. They slotted the brain into a FANUC or Yaskawa that already possessed integration, safety, maintenance, and trust. A finished-product maker must build all of that itself.

In an early-stage market, value gathers in the scarcest layer. What is scarce in Japan is neither the body (China supplies it cheaply) nor raw AI (NVIDIA and Google supply it) — it is the layer woven from local integration, safety, and trust. Foreign finished products hold none of it, so they cannot drive a wedge in.

Tellingly, Japan is now buying what it lacks: in October 2025 SoftBank Group agreed to acquire ABB's robotics business for US$5.375 billion, with Masayoshi Son declaring "the next frontier is physical AI" [4].

The lesson for foreign humanoid makers is plain: let go of "selling the machine," and enter as one body with partners that carry local safety, integration, and maintenance — designing who shoulders that layer before the product spec itself. Japan's body is strong, but only those who built it, or those who can carry local trust, get to ride it.

References

[1] "China's humanoid corps [2026 edition]," renue, Inc. https://renue.co.jp/posts/china-humanoid-robot-unitree-agibot-ubtech-galbot-dominance-2026 / "2025 global humanoid deployment volumes," Counterpoint Research, Jan. 2026. https://www.value-press.com/pressrelease/368247

[2] "NVIDIA partners with FANUC to advance physical AI," ITmedia, Dec. 2, 2025. https://www.itmedia.co.jp/aiplus/articles/2512/02/news101.html / "Majors including FANUC, ABB, and Yaskawa accelerate physical AI with NVIDIA," Robostart, Mar. 19, 2026. https://robotstart.info/article/2026/03/19/381706.html

[3] "Play Robotics founded to fill the humanoid safety-standards vacuum," Robostart, Apr. 21, 2026. https://robotstart.info/article/2026/04/21/381811.html

[4] SoftBank Group, "Notice concerning the acquisition of ABB Ltd's robotics business," Oct. 8, 2025. https://group.softbank/news/press/20251008