Korea's humanoid list has ten industries on it. A stage is not one of them

Korea's ₩2.3 trillion humanoid plan, announced August 27, sets a target of mass-production lines for humanoids specialised to ten industries by 2030. Korean press reports list them: display, aviation, logistics, retail distribution, shipbuilding, automotive, home appliances, chemicals, hospitals and hotels.

Every one of those is a labour-substitution case. A humanoid earns its place on a shipyard or a hospital ward if it costs less than the shift it replaces, and the policy's other instruments — 1,080 state-purchased units, component localisation from 45% to 80% — are calibrated to that logic. Which is why the omission is coherent rather than careless: there is no labour-cost argument for a robot that dances.

What Koreans actually paid to see

The physical AI product that has moved through Korean consumer hands this year is not a shipyard robot. Galaxy Robot Park drew a company-reported 20,000 visitors before it officially opened and began permanent operation on August 21 across a 16,500 square metre site, with robot performances, a robot arena and robots making pizza, cocktails and ice cream. The venue sells attention, not labour hours.

That is a different business, and a smaller one. It is also the only one in the country currently converting humanoid hardware into consumer revenue at any volume, which is the pattern physical AI revenue shows almost everywhere: entertainment and defence get paid first because neither requires the robot to beat a human on unit cost.

Hotels are the seam

The one category on the list that touches the same territory is hotels. A hotel humanoid is judged partly on whether guests enjoy it — a performance criterion wearing a service-sector label. If Korea's specialised humanoid programme produces anything transferable to venues, arenas and parks, it will come through that line.

The larger mismatch is architectural. Galaxy's stated next step is Soulware, an operating system meant to drive robot hardware from multiple vendors, with a robot idol in the first half of 2027 and a tour after it. An OS layer is invisible to a policy that measures success in component localisation percentages: it makes no parts, localises nothing, and can run on imported hardware. The plan's 45%-to-80% target and a software control layer are not in conflict, but the policy has no instrument that registers the second one.

The counter-case

The defensible reading is that Korea is right. Ten industrial verticals address a market orders of magnitude larger than robot performance, the country's manufacturing base is where its comparative advantage sits, and public money chasing entertainment robotics would be a subsidy to a sector already able to sell tickets. Nothing in the plan prevents entertainment operators buying the hardware it funds, and cheaper domestic hands and actuators benefit a robot idol as much as a welding humanoid.

The cost is narrower than "policy ignores entertainment". It is that consumer familiarity — the thing that decides whether households accept these machines later — is being built by a handful of private venues with no policy relationship, while the state's 1,080 robots go to universities and research institutes where the public will not encounter them.

Disclosure: Embodied Wire has a commercial relationship with Galaxy Corporation. Coverage follows the same sourcing standard as all other companies, and self-reported figures are labeled as such.