At 6 a.m., no workers are standing in the paddy field — just a drone lifting off a rooftop pad, sweeping 300 mu (about 20 hectares) of rice in 30 minutes. The "digital farm manager" has already started the day.
This is not a pilot video. It is XAG's operational 超级农场 (super farm) in Huangpu, Guangzhou, reported by People's Daily in April 2026 — and it shows what AI deployed in agriculture actually looks like when the demos end.
The field brain, not the robot
The headline is not one machine. It is a loop. Drones, BeiDou-guided tractors, and IoT sensors feed a central AI the farm calls its "field brain."
- A drone seeds nearly 300 mu (≈20 ha) in 30 minutes at 8 m/s, carrying close to 80 kg of seed — up to 50× the speed of manual sowing.
- 北斗 (BeiDou)-guided tractors prepare land with an error of no more than 2.5 cm.
- Soil sensors, weather stations, and cameras upload readings every 15 minutes; the AI turns temperature, humidity, light, and fertility into irrigation and fertilization advice.
The numbers farms actually care about
Across the Guangzhou super farm's last full cycle, XAG's operations manager told People's Daily the system:
- cut water and electricity cost per mu by 47%
- cut pesticide use by 30%
- raised fertilizer efficiency by 40%
During a rice-blast outbreak, the AI flagged zones of mild infection where reduced dosing was enough — saving more than 10,000 yuan (≈ US$1,400) in pesticide on its own. AI-planned transplanting routes also lifted land use per mu by 15%.
The Xinjiang "super cotton field" is the sharper example: two employees manage 3,000 mu (≈200 hectares), with per-mu yield 16% above conventional farming and overall cost 22.89% lower. Year-round, the farm needs only three technical staff.
From drone to robot ecosystem
On 6 July 2026, XAG held its Agricultural Robot Conference in Guangzhou and moved beyond standalone drones into an autonomous workflow (reported by Pandaily and the Guangzhou development-zone government):
- X Series drone with the SuperX 5 Apex controller (2× compute) and a 4D imaging radar plus vertical radar; power-line detection above 90%.
- XA1 docking station — millimeter-level auto-docking, 10,000 takeoff-landing cycles, automatic charging and refilling.
- LM1 smart liquid-mixing unit — handles 8 liquid sources, 60 L/min fill, one-click self-cleaning.
- B18630 smart flash battery — 2.5 minutes for a quick charge, 3.5 minutes full, ~4,000 charge cycles.
- RM80 unmanned mower — 143 kg, all-aluminum, climbs 30% slopes, covers 0.33–0.53 ha/h in orchards.
XAG is also a business: the company reported 2025 revenue above 1.166 billion yuan (≈ US$164 million), with overseas sales of 419 million yuan (≈ US$59 million), up 13% year-on-year, about 36% of total revenue, across nearly 70 countries (AgroSpectrum).
Why this is the real "AI落地" story
Healthcare and education get the headlines, but agriculture is where autonomous AI has the clearest ROI in China: labor shortage, thin margins, and huge acreage. The gain is not "cool robots" — it is fewer workers covering more land at lower input cost, with the AI doing the boring, repetitive sensing and spraying.
Honest limitations
- The Guangzhou and Xinjiang savings (47% water/electricity, 30% pesticide, 40% fertilizer; 16% yield, 22.89% cost in Xinjiang) are reported by XAG's own farm manager to People's Daily — they describe XAG demonstration "super farms," not a randomized multi-farm study.
- XAG's 2025 revenue and overseas share are company disclosures (AgroSpectrum), not independently audited by us.
- Product specs (B18630 3.5-min charge, 4,000 cycles, 90% line detection) are XAG launch claims; we found no independent lab test.
- This covers one company's ecosystem; it does not compare DJI Agras or other Chinese agtech vendors.
What readers can do now
- Run one task's ROI: if you farm or advise farms in China, price a single task — spraying or seeding — with a drone + 北斗 (BeiDou) setup versus manual labor for one season.
- Demand multi-farm data: policymakers and NGOs should visit a demo super farm but require multi-season, multi-farm yield data before scaling any subsidy.
- Plan the labor transition: automation displaces the 10–15 workers once needed per 300 mu — pair deployment with retraining, not just replacement.