Korea’s 2027 budget triples the physical AI line to ₩3.1 trillion

South Korea’s cabinet approved the 2027 budget proposal on September 1, 2026. Inside it, the line for physical AI rises from ₩900 billion to ₩3.1 trillion, a tripling, and sits within a ₩21.3 trillion package covering the government’s three mega-projects plus AI — up 97.2% from ₩10.8 trillion this year. The same package funds roughly 10,000 GPUs.

The robot spending is itemised by buyer rather than by technology. Budget-day reporting lists 650 robots for research and education, 214 for defence, 369 for police and fire services, 200 for elder care and 210 for agriculture. Those lines add to 1,643; outlets described the overall public-sector deployment as about 2,000 AI robots, so the remainder is not broken out in the day-one coverage. Separate subsidies go to manufacturers adopting humanoids and automation.

What is new versus what was already announced

Korea has been announcing physical AI money at a steady clip. In July it committed about ₩16 trillion of 2026 policy finance across six industries. On August 6 the deputy prime minister set out a plan to deploy 1,000 AI robots a year across ten industries. On August 27 the budget ministry attached a unit count to it: 1,080 domestic humanoids purchased by 2030 against ₩2.3 trillion for the humanoid ecosystem.

September 1 is the point where those intentions enter a cabinet-approved budget document. It still has to pass the National Assembly, which is where line items get moved, but the ministry-by-ministry split is now on the record.

Why it matters

Public procurement is the fastest way to manufacture demand for a domestic humanoid industry that has no commercial order book, and Korea is now doing it openly. The trade-off is price discovery. A state buyer with a unit target will pay what the plan requires; it does not tell you what the machine is worth to someone who has an alternative. That is the gap the itemisation partially closes — a police-and-fire robot and an elder-care robot are different products with different failure costs, and splitting the buy by ministry at least forces the specifications to be written down.

The second thing to watch is who supplies it. The August 27 plan set a target of raising core component localisation from about 45% to 80%, and the state has already taken equity in a robot-hand maker through the National Growth Fund. A procurement line and a supply-side stake in the same year is an industrial policy, not a purchase.