State Grid Jiangsu recently issued its “AI+” three-year action plan, built on a “2359” framework to fuse AI with grid business. Earlier, the State Grid corporation internally issued a 2026 embodied-intelligence development plan targeting about 8,500 embodied devices this year with total investment around RMB 6.8 billion, covering power inspection, live-line work, emergency rescue and warehouse logistics. Adding Southern Grid and local energy groups, industry estimates put 2026 power-sector embodied-AI investment above RMB 10 billion.
It is a procurement list to make any hardware firm’s heart race. But two bellwether financials tell a different story: YunShenChu’s first-half revenue rose 117.67 per cent to RMB 259 million yet net profit was a negative RMB 8.8 million; ShenHao signed an RMB 1.167 billion computing-centre O&M contract yet first-half net profit was still a loss near RMB 60 million. Orders are exploding; profit has not caught up. That scissors gap is the honest entry point to this power-AI shift.
Why Jiangsu, and why now
The Jiangsu plan’s weight is not the word AI but the certainty behind 2359. Two steps are explicit: build the base in 2026, lead comprehensively by 2028. Five digital bases, compute centre, data centre, model centre, intelligent hub and compute-inference platform, each needs existing informatisation assets. Jiangsu has built a provincial intelligent-compute centre at 10^17 operations per second and incubated over 2,000 agents by 2025.
Industrial structure is the real driver. Suzhou Industrial Park and Nanjing Jiangbei New Area pack dense manufacturing and dense compute demand with limited green power. Low-latency scenarios like autonomous driving and high-frequency trading must stay near users, not move west. Jiangsu also funded a power embodied-intelligence key lab with RMB 50.45 million, planning RMB 110 million over five years, aimed not at robot bodies but at a power-specialised robot team that understands electrician language, does fine work and holds the safety line. Jiangsu is racing for scenario-definition rights.
The money is in three non-robot layers
YunShenChu’s revenue rests on scenario depth, power inspection being its core. ShenHao’s RMB 1.167 billion contract splits into about RMB 327 million for robot supply and O&M and about RMB 840 million for O&M labour; first-half robot revenue was only RMB 19.39 million at 17.25 per cent margin. The data points to one fact: the power-robot track is in order-boom, profit-lag. The money is not in the robot body.
The first opportunity is the compute centre’s nervous system, the two-way signalling between compute load and grid dispatch, a high-replacement-cost moat once placed. The second is the compute centre’s security and care, inspection dogs for ten thousand substations at RMB 500,000 to 1 million each implying a thousand-billion-yuan market, where the last-mile engineering of adapting to complex electromagnetic environments is the real barrier. The third is virtual-power-plant arbitrage: data centres in Guangdong already trade in the spot market via the YueNengTou platform, and Dongguan Securities expects national VPP regulation above 20 gigawatts by 2027 and 50 gigawatts by 2030, a playground for software firms like HengHua and LangXin, not robot companies.
The structural effect
The East-West compute gap is stark: Ningxia Zhongwei’s on-grid price sits near RMB 0.36 per kWh while parts of East and South China reach 0.65 to 0.85. A 1-gigawatt IT-load data centre running full year uses about 8.76 billion kWh; under extreme price gaps the annual electricity difference could reach RMB 5 billion, though real projects are usually narrower. Zhongwei’s model already works, moving Fujian compute tasks over 2,000 kilometres to Zhongwei in minutes via China Mobile’s compute-network brain.
The body window is narrowing: the State Grid’s procurement list has already named YunShenChu, Unitree, AgiBot, UBTECH and Fourier as body suppliers. But the compute-electricity scheduling platform, O&M and VPP arbitrage are the trillion-yuan service markets. Jiangsu submitted first, but this exam has no standard answer.
Editor’s note: This is an adapted translation of the original OFweek Robotics report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-10/ART-8321202-8120-30705554.html.