A lineman in a mountain substation no longer has to walk a frozen ridge at 2 a.m. to read a gauge that could have been checked by a machine. Across China's power network, the boring, dangerous and repetitive jobs of keeping the lights on are being handed to robots — and the country's biggest utility has just turned that into a procurement plan with real numbers.
State Grid Corporation of China, the giant that moves electricity for most of the country, drew up an internal 2026 embodied-intelligence (具身智能) development plan. The headline is simple: buy robots at scale, not as toys.
The shopping list, in hard numbers
The plan targets roughly 8,500 embodied-intelligence (具身智能) devices in 2026, with total investment of about ¥6.8 billion (≈ US$0.95 billion / HK$7.4 billion). It concentrates on four scenarios — power inspection, live-line work, emergency response and warehouse logistics — with the explicit goal of moving the grid from "human-maintained" to "autonomously maintained."
Broken down by category:
- Quadruped (四足机器人) inspection dogs — 5,000 units, about ¥1.5 billion (≈ US$210 million). The single largest line item, deployed at substations, transmission lines and mountain grids.
- Humanoid (人形机器人) live-line work robots — 500 units, about ¥2.5 billion (≈ US$350 million). The highest unit price, aimed at distribution live-line work and ultra-high-voltage projects.
- Dual-arm inspection robots — 3,000 units, about ¥1.8 billion (≈ US$252 million). Focused on substation equipment operation and fault handling.
The equipment portion sums to about ¥5.8 billion, with the remainder directed to technology R&D and talent. Procurement runs in batches — pilot in Q1, scaled buying in Q3, top-up in Q4 — and devices must meet State Grid's technical specification and integrate with its "Guangming" power large model (光明电力大模型) with local deployment for data security.
Why the power grid is the natural first customer
Utilities are close to a robot-maker's dream scenario. Substations are relatively closed environments. The tasks — reading pressure gauges, judging indicator lights, recording temperatures, patrolling fences — are repetitive and dangerous. China runs more than 20,000 substations and millions of distribution rooms, and the push toward "unmanned" operation creates near-bottomless demand.
One industry estimate cited in coverage puts a巡检 dog's purchase cost at ¥300,000–500,000, against an annual per-inspector cost of about ¥300,000, with around-the-clock operation and a payback of two to three years. The safety case is sharper still: in live-line and high-altitude work, robots cut high-risk human exposure dramatically.
Who builds the dogs
The plan names a roster of robot-body suppliers: Deep Robotics (云深处), Unitree (宇树), Agibot (智元), UBTECH (优必选) and Fourier (傅利叶), among others. It also activates companies with deep grid ties — Xianheng International (咸亨国际), Ekko (亿嘉和) and Shenhao (申昊科技) — that have waited years for a procurement wave of this size.
According to Li An, chief scientist at BrainNet (脑机网), the grid is where embodied intelligence (具身智能) first stops being a demonstration and starts being infrastructure, because the scenarios are closed, repeatable and willing to pay.
From showcase to infrastructure
Past utility robot purchases were small, trial-sized orders — a handful of dogs per pilot, easy to announce and easy to forget. The scale here changes the math: thousands of a single category justifies dedicated production lines and forces component standardization down the supply chain. The local-deployment and data-security requirement also signals that the grid treats these machines as critical infrastructure, not consumer gadgets.
Analysts cited in coverage expect Southern Grid and local energy groups to follow State Grid's lead. If those follow-on buys materialize, total 2026 power-sector embodied-intelligence (具身智能) investment could push past ¥10 billion (≈ US$1.4 billion). That would make the power grid the first vertical in China to turn embodied intelligence from a demo category into a budget line.
For the workers who currently walk those ridges, the shift is double-edged. Robots take the coldest, highest and most repetitive hours, but they also change what a grid job looks like: fewer boots on frozen gravel, more eyes on screens and exception handling. That reframing — maintenance as monitoring — is where the real cultural work of adoption lives, and it is rarely captured in a procurement headline.
Honest limitations
The figures come from a plan described in reports as "internally issued" by State Grid, not a public, legally published procurement document. They are planning targets, not signed contracts, and the exact mix and timing can shift. The ¥300,000–500,000 per-dog and payback estimates are industry commentary, not audited results. Independent, on-the-ground verification of deliveries is limited so far. Read this as a strong directional signal of scaled utility adoption, not a guaranteed spend.
What readers can do now
- If you work in energy or industry: watch whether Q3 "scaled procurement" actually converts to deliveries — that batch, not the announcement, is the real test.
- If you follow robotics supply chains: track the named body suppliers and the grid-tied specialists; a utility order book is a rare, fundable demand signal.
- If you evaluate deployments: separate the safety narrative from measured outcomes — ask for fault-handling time and incident data, not just demo videos.