Fujitsu enlists FANUC, Yaskawa and Kawasaki to build a sovereign physical AI platform on NVIDIA technology
Fujitsu said on July 16 it had begun exploring physical AI businesses with three of Japan's largest robot makers, FANUC, Yaskawa Electric and Kawasaki Heavy Industries, using NVIDIA Cosmos, Omniverse, Isaac and the Newton physics engine. Fujitsu will build a common software platform and hardware interfaces, with manufacturing, retail and logistics, and healthcare as first targets. No timeline or financial terms were given.
Fujitsu Ltd. said on July 16, 2026 it had begun exploring physical AI businesses with FANUC Corp., Yaskawa Electric Corp. and Kawasaki Heavy Industries Ltd., bringing three of Japan's largest industrial robot makers into one initiative built on NVIDIA technology. Fujitsu will lead the effort and develop the software platform and hardware interfaces that the partners would share, according to its release.
The agreement is at the exploration stage. Fujitsu said it would draw up technology development and business expansion roadmaps with the participating companies but gave no timeline, investment amount or revenue target. Each robot maker announced its own track with Fujitsu on the same day, and the releases describe parallel bilateral discussions rather than a joint venture.
The technology stack
Fujitsu named four NVIDIA technologies: the Cosmos world foundation models, the Omniverse simulation platform, the Isaac open robotics platform and the Newton physics engine. Fujitsu's own contribution is an autonomous AI platform that combines its Takane large language model with NVIDIA technology, and a stated aim to build a sovereign collaborative control platform for Japan and open it as shared infrastructure. FANUC brings robots that support open platforms through ROS 2 and Python, Fujitsu said.
The release lists three application areas. In manufacturing, the partners want systems that optimise production planning and let factories adapt autonomously on site. In retail and logistics, they target material handling driven by real-time sales and inventory data. In healthcare, they aim to automate the transport of pharmaceuticals and specimens and outpatient reception services.
What each robot maker said
FANUC said in its release that it had shown its first physical AI system at the International Robot Exhibition in December 2025 and that current applications include robot operation by voice recognition and component kitting by an AI agent. FANUC President and CEO Kenji Yamaguchi called the collaboration an important step toward significantly advancing and accelerating the real-world deployment of physical AI powered by robotics. Fujitsu President and CEO Takahito Tokita said the aim was to connect factory robots, equipment, people and business applications to optimise production and run it autonomously.
Kawasaki Heavy Industries framed its track around hospitals. Its release said the two companies would combine hospital information systems, robots and AI, with Kawasaki developing autonomous task execution for medical transport and AI agents for patient services. Kawasaki President and COO Yasuhiko Hashimoto said the goal was a one-stop hospital solution covering the patient journey from arrival to post-operative care. Kawasaki also noted that it opened the Kawasaki Physical AI Center San Jose in May 2026.
The companies by the numbers
The four Japanese partners reported mixed results in their latest quarters. Fujitsu's revenue rose 3.9% to ¥779.4 billion in April to June 2026 and operating profit jumped 56.9% to ¥52.5 billion. FANUC's net sales rose 17.7% and operating income 26.1%. Kawasaki's revenue grew 11.3% and business profit 74.3%. Yaskawa, whose quarter ran from March to May, grew revenue 10.6% but saw operating profit fall 19.2%, which its filing attributed in its robotics segment to production effects from a core system migration and restructuring costs in Europe.
- Fujitsu
- Lead, platform
- Apr to Jun 2026
- 779,400
- 3.9
- 52,500
- 56.9
- FANUC
- Robot maker
- Apr to Jun 2026
- 231,035
- 17.7
- 53,492
- 26.1
- Kawasaki Heavy Industries
- Robot maker, healthcare
- Apr to Jun 2026
- 543,576
- 11.3
- 35,752
- 74.3
- Yaskawa Electric
- Robot maker
- Mar to May 2026
- 138,982
- 10.6
- 8,486
- -19.2
Figures as disclosed in quarterly earnings filings. Profit is operating profit for Fujitsu, FANUC and Yaskawa and business profit for Kawasaki. Fujitsu figures disclosed as ¥779.4 billion and ¥52.5 billion. Yaskawa's fiscal year ends in February. NVIDIA is a technology partner and is not shown.
As of Oct 1, 2026
The initiative is Japan's answer to a pattern already visible in the United States and China, where AI platform companies and robot makers are pairing up. For the robot makers the attraction is access to simulation and world models that none of them would build alone. For NVIDIA it secures the leading Japanese arm makers as users of Cosmos, Omniverse and Isaac. Fujitsu's role is to sit between them as the integrator and to offer a domestically controlled control layer, which it calls sovereign.
In plain terms, world foundation models such as Cosmos generate realistic synthetic video and physics so that robot policies can be trained and tested in simulation before they touch real equipment. Omniverse and Isaac provide the digital twin and robot simulation tools, and Newton models contact and physics. Fujitsu's interfaces are meant to let models trained this way control robots from different makers through one platform.
The open questions are significant. The three robot makers compete fiercely, and none has said it will share data or controller access with the others. The term sovereign sits awkwardly with a stack whose core simulation technology comes from a U.S. company. There is also overlap: FANUC and Yaskawa have since announced separate AI work with Google, and FANUC signed a partnership with Hitachi on Sept. 30.
The next markers are the roadmaps Fujitsu promised, any first joint demonstration, the companies' results for the July to September quarter, and the International Robot Exhibition, where Japanese makers traditionally show new systems.
ROBOTNESS analysis
Fujitsu's initiative is more valuable as a standards play than as a business, and its success depends on whether rival robot makers accept a shared control interface.
The evidence is in the structure. Fujitsu has three separate bilateral discussions rather than a consortium with governance, a budget or a shared entity. Its defined deliverable is a common software platform and hardware interfaces, which is the layer that becomes valuable only if several makers adopt it. The NVIDIA components give all three a common simulation base.
The strongest counter-argument is that FANUC, Yaskawa and Kawasaki have every reason to keep their controllers proprietary, and in the months since July each has pursued separate AI partners. A shared interface could commoditise the very control software that differentiates them.
Bull case: Fujitsu publishes an open interface specification, at least two robot makers ship support for it, and Japanese hospitals and retailers buy integrated systems from Fujitsu with robots from several makers. Fujitsu gains a recurring platform business and Japan a domestic alternative to foreign robot stacks.
Bear case: the discussions stay bilateral, each robot maker builds its own AI layer with Google, Hitachi or NVIDIA directly, and Fujitsu's platform becomes one of several options rather than the standard. The initiative then yields demonstrations but little revenue.
Signals to watch:
- By March 31, 2027 (end of Fujitsu's fiscal 2026): publication of the promised roadmaps or an interface specification.
- December 2027: the next International Robot Exhibition in Tokyo, the natural venue for a joint demonstration if the biennial schedule holds.
- Quarterly results in late October and early November 2026: whether any of the four companies mentions physical AI orders linked to the initiative.
- Fujitsu
- Business exploration; roadmaps to follow
- July 16, 2026
- FANUC, Yaskawa Electric, Kawasaki Heavy Industries
- Hospital transport and patient service AI agents
- Not disclosed
- Manufacturing, retail and logistics, healthcare
- Cosmos, Omniverse, Isaac, Newton physics engine
- Common software platform, hardware interfaces, Takane LLM
Why it matters
This is the first time three of Japan's largest robot makers have aligned, even loosely, around a common AI and simulation stack. Japan's robot industry has competed on controllers and reliability, and none of the three has the scale in AI models that U.S. and Chinese platform companies are building. A shared base would let them pool demand for simulation and synthetic data instead of each paying separately.
For Fujitsu, physical AI offers a route back into hardware-adjacent revenue. Its Q1 operating profit rose 56.9%, but full-year revenue is forecast to grow only 0.2%, and a platform role in robotics would add a new growth line.
Rival analysis
The comparable moves elsewhere are bilateral. FANUC has worked with Google since May and with Hitachi since September. Yaskawa has Google DeepMind and SoftBank. Hitachi is building its own HMAX physical AI portfolio and edge chip, which makes it a direct rival to Fujitsu as an integrator for FANUC.
Internationally, NVIDIA offers the same Cosmos, Omniverse and Isaac stack to every robot maker, so the Fujitsu initiative does not give Japanese makers exclusive technology. The differentiation must come from Fujitsu's integration, Japanese-language models such as Takane, and domestic data control.
Valuation context
No money was committed, so the initiative does not affect estimates directly. The four Japanese partners together reported first-quarter revenue of about ¥1.69 trillion (sum of the figures in the table), with Fujitsu and Kawasaki the largest.
The market-relevant read is on profitability. Fujitsu and Kawasaki are expanding margins, FANUC is growing strongly, and Yaskawa's 19.2% fall in operating profit suggests it has the least room to fund a platform effort.
Supply-chain implications
The NVIDIA stack implies GPU compute for simulation and training, which Fujitsu can supply through its data centre business. At the robot level, healthcare transport and outpatient services would rely on mobile robots and service robots more than on industrial arms, a segment where Kawasaki has been building presence.
If Fujitsu's hardware interfaces become common, controller suppliers and sensor makers would need to support them, which would lower switching costs between robot brands for end users.
Signals to watch
Publication of Fujitsu's roadmaps and any interface specification; a first multi-vendor demonstration; any hospital or retail customer named by Fujitsu and Kawasaki.
Also watch whether FANUC and Yaskawa reference the Fujitsu platform when they announce new AI products, which would indicate real adoption rather than parallel strategy.
Analyst view
Thesis: the initiative will matter only if it produces a shared interface that at least two of the three robot makers adopt; otherwise it remains a set of separate pilots. Confidence: low.
Reasons for low confidence: no governance, budget or timeline has been published, and each robot maker has since announced competing AI alliances. The participation of all three leaders and NVIDIA keeps the upside open.
Questions you should be asking
Will Fujitsu publish an open interface specification, and will the robot makers commit to supporting it in their controllers? Who owns data generated in joint deployments?
How does Fujitsu reconcile a sovereign platform with dependence on NVIDIA's simulation stack? Will Hitachi, a rival integrator, join or compete?