# KDDI and ELYZA Robotics: The 30 Billion Yen Physical AI Plan

KDDI and ELYZA announce a robotics company and FY2030 investment and deployment targets, with practical work starting through proofs of concept.

By Paras Madan | 2026-10-28


KDDI and ELYZA announced ELYZA Robotics on October 27, with 30 billion yen in planned research spending through fiscal 2030 and a target of more than 1,000 deployed robots. The effort puts workplace implementation alongside model research; those installations are a future goal.

The company began operating on October 1, before the public announcement.

## What KDDI plans to build

The [joint announcement](https://newsroom.kddi.com/news/detail/kddi_nr-1209_4783.html) targets more than 300 physical-AI specialists by fiscal 2030, working on robot models and deployment platforms.

The group plans to procure industrial, quadruped and humanoid robots from outside suppliers. Proofs of concept start from 2026, targeting retail, logistics and manufacturing.

KDDI describes separate research and operating platforms. The former supports model development and field-data collection; the latter supports functions such as distributed inference and remote monitoring. The announcement does not supply a general commercial price list, public model weights or completed customer results for this new programme.

For a potential partner, that positions ELYZA Robotics as a development and integration effort. It does not yet identify a single standardized robot product that every customer can order and deploy unchanged.

## The Sakai test environment

A separate [Alliance Park announcement](https://newsroom.kddi.com/news/detail/kddi_nr-1209_4784.html), also dated October 27, describes a practical development setting in Sakai, Osaka. KDDI plans to bring together model developers, robotics businesses and research organizations, with ELYZA Robotics machines introduced in stages.

The proposed loop connects operations with training: run robots and other autonomous systems at the site and in nearby industrial settings, collect data, use it to improve the technology, then return improvements to the field. This is a plan for building evidence through deployment, not a published result showing that the loop already improves performance.

For example, a shelf-stocking pilot could reveal that a robot handles an object reliably in a clear test area but struggles when packaging is damaged or a person blocks its route. Keeping those failures connected to the exact task and conditions is useful for deciding what to improve. That example illustrates the work ahead; it is not a reported ELYZA result.

## Why the deployment target needs context

A count of installed robots tells only part of the story. A pilot machine operating occasionally and a machine completing a full shift contribute equally to a unit count, while providing very different evidence of usefulness.

The more informative milestones would include the tasks deployed, operating hours and frequency of human intervention. For buyers, the time required to integrate a robot with existing equipment and train staff can matter as much as the model's headline capability.

KDDI's plan is notable for funding the operational layer alongside AI development. The next useful updates will show how the proofs of concept become repeatable deployments, which responsibilities stay with a site operator and what support is needed when a machine cannot complete its task. Those details will make the fiscal-2030 target easier to judge.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders explain AI platforms and deployment plans in clear content that developers, partners and buyers can evaluate.

[Book a 30-minute call](https://cal.com/paras-madan/30min).