South Korea funds about 69% of Doosan Robotics' 98.9 billion won plan for AI chips and nuclear welding robots

Of the 98.9 billion won, 68.1 billion is government subsidy routed through the Industry Ministry and 30.8 billion is Doosan's own money ^[3][11]. That puts the state's share at about 69%, or roughly 2.2 won of public money for every won Doosan commits ^[12][13]. The totals cover both deals, the chip-and-robot program and the welding cell, and the budget for each was not broken out ^[1][7].

Welding is the shorter bet. Its 45 months run nine months under the four and a half years given to Doosan's on-device chip ^[14]. If the 50% cut in production time held across a whole job, one cell would turn out twice the welds in the same hours ^[17]. That figure is a target. The first real measurements come from the trials at Doosan Enerbility's nuclear equipment plants ^[2].

In one version, the cell passes verification on schedule ^[6], moves into shipbuilding and defense ^[16], and the nuclear certificate becomes Doosan's sales credential in two more industries ^[2]. In another, the chip is the product and welding is where it gets tested. DeepX described the project in close to those terms. "This project will be a concrete case of physical AI, in which ultra-low-power AI chips are combined with robots and used at actual manufacturing sites," DeepX said ^[8].

In a third, DeepX's neural processing unit and Doosan's own chip end up competing for the same on-device job. DeepX builds the welding cell's edge AI hardware on a domestic neural processing unit, plus a three-channel vision pipeline and thermal-imaging models ^[5]. Doosan's other project develops its own system-on-chip with a neural processing unit and real-time control ^[7]. By the 2031 date set for its new cobots and humanoid, Doosan expects to sell a chip of its own built for the same on-device work ^[7].

Doosan's spending follows its chief executive's diagnosis. "Industrial labor shortages and aging skilled workforces cannot be solved by scaling up traditional automation," Doosan Robotics CEO Park In-won said ^[9]. So the company's 30.8 billion won goes into domestic chips and cobots ^[1][3]. Processing stays on the robot, without cloud servers or PCs, to remove latency and security risk at nuclear and defense sites ^[10].

I think the welder shortage ^[4] is the case for the grant, and until commercial sales start in 2030 the paying customer is the state ^[3][6]. If the Enerbility trials come near halving production time on nuclear parts, the shortage becomes a reason for shipyards to pay full price, and 68.1 billion won will have been a cheap way for the state to get there ^[3][15][16]. A purchase order from a shipbuilder or defense buyer before 2030 would prove me wrong ^[6][16].