Doosan Robotics is developing its own AI chips to power what the Korean company calls "Korean Physical AI" — an effort to embed locally designed intelligence into the robots it builds, Korea IT Times reports.

The move places Doosan in a small but growing group of Korean manufacturers designing semiconductors in-house rather than buying general-purpose processors from U.S. vendors. For a collaborative-robot (cobot) maker, custom silicon means tighter control over how machine vision, sensor fusion and motion planning run on the device itself, without depending on cloud connectivity or external accelerator suppliers.

What is 'Physical AI'?

The term refers to AI systems that act in the physical world — robots, humanoids, autonomous machinery — as opposed to chatbots and other software-only applications. It has become a rallying point for Korean industry and government, which see the country's manufacturing base as a natural advantage in this segment.

Doosan Robotics is one of Korea's leading cobot makers. By pairing its robot platforms with homegrown AI chips, the company aims to differentiate on integration: hardware, control software and inference silicon designed together, rather than assembled from third-party components.

Why in-house chips?

The Korea IT Times report does not disclose process nodes, wafer volumes, chip specifications or a production timeline. What it does establish is strategic direction: Doosan wants Korean-designed AI semiconductors inside Korean-built robots, a positioning that serves both commercial and industrial-policy goals.

For robot OEMs, the logic of custom silicon follows a familiar pattern:

• Latency and power. On-device inference avoids round-trips to the cloud, critical for real-time robot control and battery-powered platforms.
• Supply security. Domestic chip supply reduces exposure to export controls and allocation battles for foreign accelerators.
• Differentiation. Proprietary silicon is harder for competitors to replicate than robots assembled from catalog parts.

Korea's national strategy amplifies the commercial case. The government has pushed development of Korean AI chips — spanning NPUs for data centers and edge devices — as part of a broader effort to reduce dependence on foreign semiconductor IP. Doosan's initiative extends that push from digital AI into the physical realm of robotics.

Competitive context

Doosan is not alone in linking robotics with custom compute. Global robot and automation vendors are increasingly evaluating their own inference silicon as AI workloads move from pilot projects into production lines. The question for Doosan is whether a cobot maker has the volume to justify chip development costs that the industry typically spreads across millions of units.

The Korea IT Times report frames Doosan's push as a bet on that equation changing — on robots with embedded AI becoming high-volume products as manufacturing, logistics and service sectors automate.

What to watch

Key details remain undisclosed: which Korean foundry or design partner would produce the chips, what process node is targeted, and when silicon would appear in shipping robots. Doosan has also not said whether the chips will remain internal or be sold to other robot makers.

Until those specifics emerge, Doosan's announcement reads as a strategic commitment rather than a product launch. If the company names a fabrication partner and a timeline, "Korean Physical AI" becomes a measurable silicon program; without them, it remains a roadmap. The competitive stakes, however, are clear — the first robot makers to pair reliable on-device AI with affordable domestic chips will set the reference architecture for the physical-AI market Korea is racing to lead.

This article is based on a Korea IT Times report; the report did not include capacity, node or financial figures.