Korea's physical-AI push explained: the alliances, the programmes and who is getting the money
Korea funds physical AI through at least four ministry programmes and several public-private alliances, and the largest awards disclosed so far have gone to established manufacturers rather than startups.
Of KRW 444.9 billion in project budgets that MEKAWIRE has verified from company and institution releases, Daedong leads KRW 297.7 billion, Doosan Robotics KRW 98.9 billion and Neuromeka KRW 17.5 billion, and no project is led by a venture-backed startup; startups enter as consortium members, chip suppliers or through regulatory sandbox approvals.
Korea's physical-AI money comes through at least four ministry channels and is reaching established manufacturers first. Of KRW 444.9 billion in project budgets disclosed since September and checked by MEKAWIRE against the companies' and institutions' own releases, two-thirds sits in projects led by the farm-machinery maker Daedong, a fifth with Doosan Robotics, and none in a project led by a venture-backed startup.
That matters because the government's own targets are about breadth. When the trade ministry launched the K-Humanoid Alliance in April 2025 it spoke of about 40 founding members and more than KRW 1 trillion of public and private investment by 2030. The question this explainer answers is where the first concrete awards have landed, and what that says about the kind of industry Seoul is building.
Six projects, three companies and a university network
- Daedong
- National agricultural AX platform
- 254.6
- n/d
- n/d
- Sept 7, 2026
- Doosan Robotics
- Two KEIT-run projects: AI-chip cobots and humanoid; nuclear-component welding (MOTIR)
- 98.9
- 68.1
- 54 and 45 months
- Oct 6, 2026
- Daedong
- K-On-Device AI Semiconductor, unmanned farm-robot track (MOTIR, KEIT)
- 43.1
- n/d
- Jul 2026 to Dec 2030
- Sept 9, 2026
- Five universities incl. Ajou (Army hub)
- Defence AX hubs (MND, IITP)
- 19.0 (2026)
- n/d
- 2026
- Oct 13, 2026
- Neuromeka
- Human-AI collaborative action model for ship welding (MSIT)
- 17.5
- 10.5
- Aug 2026 to Dec 2030
- Sept 28, 2026
- Seil Forging (MakinaRocks co-developer, KRW 2.4bn)
- Machinery programme, hot-forging AI (MOTIR, KEIT)
- 11.8
- n/d
- Jun 2026 to Dec 2029
- Oct 14, 2026
Company and institution releases verified by MEKAWIRE. Disclosed figures only, not estimates; n/d = not disclosed. Total 444.9; Daedong-led 297.7 (66.9%).
As of Oct 16, 2026
Daedong is the largest single recipient in our data. It said in September that it leads the unmanned farm-robot track of the K-On-Device AI Semiconductor programme, a KRW 43.1 billion project with 14 partners that runs to December 2030, and earlier that it leads the KRW 254.6 billion national agricultural AX platform. Doosan Robotics said on Oct 6 that two projects it proposed to KEIT-run programmes were selected, with total budgets of about KRW 98.9 billion, KRW 68.1 billion of it government money. Neuromeka leads a KRW 17.5 billion science-ministry project, KRW 10.5 billion from the state, to train an action model that welds curved ship-hull blocks. MakinaRocks is a co-developer in a KRW 11.8 billion forging project led by Seil Forging, with KRW 2.4 billion for itself. The defence ministry's five AX hubs, one of which Ajou University opened for the Army on Oct 12, carry KRW 19 billion this year.
Formula: Daedong share = (254.6 + 43.1) / 444.9 = 66.9%. Government share where both figures are disclosed = (68.1 + 10.5) / (98.9 + 17.5) = 67.5%.
The pattern is consistent: the lead organisation is a listed manufacturer or a university, and the money follows an existing customer base. The next question is what the umbrella programmes above these projects are for.
The umbrella programmes are buying a Korean stack, not robots
- Regional physical-AI programme (North Jeolla, South Gyeongsang)
- MSIT
- KRW 1,413.1bn
- 2026 to 2030
- K-On-Device AI Semiconductor
- MOTIR, KEIT
- About KRW 800.2bn for ten customer-specific AI chips
- To 2030
- Robot R&D and field demonstrations
- MOTIR
- About KRW 180bn R&D and KRW 76bn demonstrations
- 2026
- K-Humanoid Alliance
- MOTIR (then MOTIE)
- About 40 founding members; over KRW 1tn public and private investment expected
- Apr 2025 to 2030
- M.AX alliance, AI robot division
- MOTIR
- About 280 members; humanoid pilots with SK Energy, HD Hyundai Heavy Industries, BGF
- Results May 2026
- Physical AI Standardization Alliance (PASA)
- MSIT ICT standards programme
- About 20 member companies
- Since Aug 2026
- Regulatory sandbox
- MOTIR
- Robros 20 humanoids at four sites; Maum AI quadruped on public paths
- Approved Oct 13, 2026
Ministry releases and earlier MEKAWIRE reports from primary sources; PASA membership as reported by TechWorld citing KOSA.
As of Oct 16, 2026
Read the project descriptions together and one requirement repeats. Doosan's 54-month project puts a Korean AI system-on-chip with a neural processing unit into three cobots and a two-armed humanoid, with Mobilint and AIDIN Robotics as partners and DEEPX on the second project. Daedong's track exists to put a Mobilint-designed farm AI chip on its robots. The K-On-Device AI Semiconductor programme behind it carries about KRW 800.2 billion to 2030 for ten customer-specific chips. The science ministry's regional physical-AI build-out in North Jeolla and South Gyeongsang commits KRW 1.4131 trillion from 2026 to 2030.
MEKAWIRE analysis: the state is funding sovereignty first and humanoids second
Korea's physical-AI programmes are an industrial-policy bet on domestic chips, data and integrators, with humanoids as the showcase; the early money therefore flows to firms that can host a supply chain, not to the startups that build the robots.
The evidence is in who leads and what they must use. In every company-led project above, the lead already sells hardware at volume (tractors, cobots, forgings) and the deliverable includes Korean silicon, Korean data or a Korean action model. The alliances reinforce that: the trade ministry's M.AX AI robot division had about 280 members in May and runs humanoid pilots with SK Energy, HD Hyundai Heavy Industries and BGF, and the K-Humanoid Alliance set 2028 targets for a shared robot foundation model and humanoids under 60 kg with more than 50 degrees of freedom.
The strongest counter-argument is that startups are getting something more valuable than grants: permission. On Oct 13 the trade ministry's sandbox let Robros run 20 humanoids next to workers at Nepes, Sungwoo Hitech, HD Hyundai Samho and Delivery Lab and let Maum AI patrol public riverside paths with a quadruped. Neither approval came with a budget, but both create reference sites that grant money cannot buy.
Bull case: the chip and data requirements produce a Korean robot stack that Doosan, Daedong and Neuromeka export by 2030, and startups ride it as suppliers. Bear case: money spread over dozens of consortia in four ministries buys prototypes and certificates, while Chinese makers win the price war on hardware before Korean pilots reach production.
What would change our view: a project of KRW 30 billion or more led by a venture-backed robot startup, or a humanoid pilot converting to a purchase order with a published unit count.
Signals to watch
The National Assembly forum on AI and robot competitiveness on Oct 21; Robotworld 2026 from Nov 4, where several of these consortia are due to show prototypes; and the 2027 budget bill, which will show whether the MOTIR robot line (about KRW 180 billion of R&D and KRW 76 billion of demonstrations in 2026) grows or is folded into the physical-AI envelope.
- Leads the KRW 254.6 billion national agricultural AX platform and the KRW 43.1 billion K-On-Device AI chip farm-robot track (Jul 2026 to Dec 2030, 14 partners)
- Oct 13, 2026: Robros may deploy 20 humanoids at four sites; Maum AI quadruped patrols on public riverside paths
- K-Humanoid Alliance (Apr 2025, about 40 founders, more than KRW 1 trillion investment expected by 2030); M.AX AI robot division about 280 members (May 2026)
- KRW 17.5 billion MSIT project, KRW 10.5 billion government, for a ship-hull welding action model (Aug 2026 to Dec 2030)
- Two KEIT-run projects, KRW 98.9 billion total, KRW 68.1 billion government funding (announced Oct 6, 2026)
- MSIT regional physical-AI programme KRW 1.4131 trillion (2026 to 2030); K-On-Device AI Semiconductor about KRW 800.2 billion to 2030
- KRW 444.9 billion across six projects verified from company and institution releases (Sept 7 to Oct 14, 2026)
Why it matters
Korea's targets are about breadth (a trillion won of investment, hundreds of alliance members, 2028 humanoid specs), but the first verified awards concentrate in a few incumbents. Investors valuing Korean robot startups on the assumption of state support should separate permission (sandbox approvals, pilot sites) from cash, which so far goes to project leads with manufacturing scale.
The chip requirement inside the largest projects also creates a captive early market for Korean NPU designers such as Mobilint and DEEPX, which may matter more for that segment than for the robot makers.
Rival analysis
Among Korean robot makers, Doosan Robotics now has the clearest state-funded humanoid path (a two-armed omnidirectional industrial humanoid on a Korean SoC), while Neuromeka has positioned for shipbuilding welding and Rainbow Robotics appears as a platform partner in Daedong's farm track rather than as a lead. Startups such as Robros and Maum AI have reference deployments without project budgets.
Against China, where city funds back humanoid makers directly with equity, Korea's model routes money through consortia led by end-users and suppliers, which is slower to produce unit volume.
Valuation context
Grant funding is not equity and does not change company valuations directly, but it lowers R&D cost: in Doosan's case the state covers about 69% of the budget. For listed leads (Doosan Robotics, Daedong, Neuromeka, MakinaRocks) the awards are small relative to market value, so share-price effects should be read as sentiment rather than earnings.
For private startups the relevant comparison is venture rounds: WIRobotics raised KRW 95 billion in May, almost as much as Doosan's two projects combined.
Supply-chain implications
The projects pull demand toward Korean NPUs (Mobilint, DEEPX), force-torque sensing (AIDIN Robotics in Doosan's consortium), and domestic integrators. Reducers, high-end servo drives and many sensors remain imported, and none of the disclosed projects names a domestic reducer supplier, which leaves the hardware bill of materials exposed to Japan and China.
Signals to watch
Oct 21: National Assembly forum on AI and robot competitiveness. Nov 4 to 7: Robotworld 2026, where consortia may show prototypes. December: National Assembly vote on the 2027 budget, showing whether robot lines grow or move into physical-AI envelopes. Any humanoid pilot (SK Energy, HD Hyundai Heavy, BGF, Robros sites) converting to a purchase order.
Analyst view
Thesis: Korea is funding a sovereign stack first and humanoids second. Confidence: medium. The project data are consistent but small (six projects), and several large programmes have not yet published award lists, which could still show startup-led projects.
Questions you should be asking
For Doosan Robotics: what share of the KRW 68.1 billion flows to partners, and when does the Korean SoC cobot reach a priced product? For Daedong: how is the KRW 254.6 billion AX platform split between Daedong and partners? For the ministries: will any physical-AI award above KRW 30 billion be led by a venture-backed company?