Kawasaki, Softbank, Sony, Honda and 40 other companies partnering to build Japanese physical AI

It's probably news to just about everyone, but Kawasaki Heavy Industries has been building humanoid robots since 2015. The ninth-generation Kaleido stands 190 cm tall, weighs 99 kg, carries 30 kg of payload, and moves at 5 km/h ... serious specs from a company with serious industrial credibility. 

What Kaleido has lacked is a path to genuine autonomy. A newly announced partnership with Noetra is Japan's answer to that problem.

Noetra is a Tokyo-based physical AI startup building an omni-modal foundation model for robots. What makes it unusual is who is behind it: not a small venture team, but a consortium of 44 Japanese companies — including SoftBank, Sony, NEC, Honda, and Kawasaki among them — that have collectively decided Japan needs a domestically developed physical AI stack before a foreign competitor builds the one everyone else is forced to use.

The partnership makes Kaleido the first planned adopter of Noetra's technology. 

The targets and not incredibly bold, however: initial deployment in controlled industrial environments by 2030, full autonomy in unstructured environments by 2040.

Possibly, they're realistic?

Japan's industrial consortium model

Japan's most durable technology industries, whether automotive, consumer electronics at their peak, semiconductors during their ascent, were often built through coordinated industrial strategy rather than pure market competition. The Noetra consortium of 44 companies follows a similar model in physical AI: pool the corporate resources and market distribution of established players, fund the R&D through a purpose-built startup, and establish a foundation model standard before the US or China do it for you.

(Well, they might be a little late on that last part.)

The structure matters here. Essentially, the foundation model race in physical AI is a standards race. Whichever models become defaults for robot training gain compounding leverage: more deployments generate more data, which improves the model, which enables more deployments. It's similar to a platform advantage in desktop or mobile computing. While OpenAI and Anthropic have that position in language AI, we don't yet have clear leaders in physical AI. Japan, through Noetra, is betting a coordinated domestic consortium can get there, at least for a domestic standard.

Noetra's omni-modal foundation model is targeted for 2028. 

The "omni-modal" suggests a model designed to process multiple input types simultaneously — vision, force feedback, language, proprioception — which is precisely what deployed industrial robots actually need.

The NVIDIA connection and the US hub

Alongside the Noetra partnership, Kawasaki is establishing the Kawasaki Physical AI Center in San Jose, with NVIDIA, Analog Devices, Microsoft, and Fujitsu as partners. That's a deliberate transatlantic structure: Japan's domestic AI stack (Noetra) paired with American simulation infrastructure and compute (NVIDIA's Isaac platform), with an initial focus on healthcare and elder care.

NVIDIA's involvement is expected at this point: the company has made physical AI a primary growth vector and is partnering with most serious players in the space. But the combination of a 44-company Japanese consortium, a US innovation hub, 11 years of humanoid hardware development, and a foundation model on a defined roadmap is an unusually complete (and complex) ecosystem relative to most humanoid startups starting from scratch.

Early deployment targets are facility patrols and parts transport, with product assembly to follow. Kawasaki's stated fleet ambition: tens of thousands of robots by 2040. Target industries are automotive and semiconductor manufacturing: exactly the sectors where Japan has the deepest existing customer relationships.

The timeline problem

The goals are 2030 for controlled environments and 2040 for unstructured autonomy.

I guess Kawasaki isn't overpromising. And maybe they're realistic, who knows. But they're kind of the opposite of aggressive. And they're long enough that the competitive landscape will look unrecognizable when those targets arrive. Figure, Agility, and their Chinese counterparts are on 2025-2027 deployment timelines. Companies racing to deploy now will have years of operational data, trained models, and locked-in customer relationships before Noetra's foundation model ships in 2028.

Arguably, 2030 and 2040 are too late.

Also arguably, if work proceeds faster, Kawasaki and Noetra will ship faster.

Japan's bet is that the quality of the foundation model, combined with the industrial depth and distribution reach of 44 corporate backers, will matter more long-term than early-mover speed. Whether that plays out like automotive — where quality and reliability win over time — or like smartphones where platform lock-in wins early and never lets go is the question the market will eventually answer.

Kawasaki has been patient with humanoid robots for 11 years. It, apparently, is continuing to be patient with physical AI. 

Whether 2040 is patient ... or just late ... is something we'll know in a few years.

Sources

• Interesting Engineering — Kawasaki bets on physical AI to make Kaleido humanoid fully autonomous
• Nikkei Asia — Kawasaki Heavy aims for fully autonomous humanoid AI robot by 2030
• Nikkei Asia — Kawasaki Heavy to partner with Nvidia on physical AI, open US robot center
• Digitimes — Kawasaki teams with Noetra on fully autonomous humanoid robot