State farms and provincial deployment vehicles — Beidahuang, Heilongjiang province, and the thinnest XPCC evidence in the corpus

East-Asia (China — Heilongjiang province and the Beidahuang reclamation area; a shorter XPCC/Xinjiang section)

Content

The September 2026 China cycle established the corpus’s reserve-and-platform-first reading of Chinese agri-AI and left two named leads open: the agricultural LLMs (written as units/china-agricultural-llms-landscape.md) and the state farms and provincial deployment vehicles. This unit writes the second. Its substance: Beidahuang is the corpus’s largest single agri-AI deployment vehicle anywhere — a state farm group that owns roughly 48M mu of farmland, runs its own information company, its own satellite-analytics estate, its own service-fee platform, and a joint state key lab with HIT and NEAU — and Heilongjiang province is the first province to issue a smart-agriculture implementation plan through the provincial government office, with Beidahuang and Jiamusi designated the province’s two smart-agriculture 样板区 (model zones) in 2025. The XPCC, by contrast, is the thinnest evidence in the corpus’s China coverage: division-level cases exist; no corps-level AI or smart-agriculture programme document was found (G-479).

Beidahuang — the data-platform estate (company newsroom, 21 April 2026, read in full)

Beidahuang Information Co. (北大荒信息有限公司), the group’s smart-agriculture subsidiary, reports an estate of:

• Platforms: 智壤 coomap (the “one map”), 离朱 intelligent remote-sensing cloud, 耘图 large-scale planting-management platform, and the 河洛 platform in iteration — 99 algorithms integrated, ~3,000 datasets online, crop-growth algorithms refreshing across the reclamation area within 5 days, auto-generating plant-protection prescription maps and ~50 custom remote-sensing report types.
• Scale: 48,000+ GIS layer services; 19 agricultural systems fully integrated; 48 satellites completing reclamation-wide staple-crop monitoring in 2 hours; yield-estimation and farm-progress monitoring accuracy improved 5-10%; pilot-farm seedling recognition >87%; smart-equipment platform covering 111 farms with 84,000 connected units; sensing across 4.64M mu; smart irrigation on 54 farms with 19,000 automated gates (water saving 10-20%; a 1,000-mu rice block saving CNY 12,000/year in water management); smart fertilisation at −10-20% fertiliser, rice +3-5% yield, +80-100 CNY/mu income.
• Recognition: first-batch National Data Administration public-data application demonstration scenario; first prize, Heilongjiang 数据要素× contest (modern-agriculture track) and second prize at the national finals; first prize, national digital-village innovation contest; selected into MARA’s 2025 smart-agriculture promoted technologies.
• Unmanned farms (overlapping units/china-ground-robotics-unmanned-farms.md, which holds the robotics detail): 万亩-scale operations with paddy/dryland labour reduced 50.3% / 51.5%; the “北大荒未来农场” model piloted at Bawuwu and Heshan farms over 15,000 mu (Bawuwu soybean +5% yield, −5% fertiliser), and written into the provincial implementation plan as a promoted technology.
• R&D: 127 research projects (46 national key, 45 provincial/ministerial); seven landmark outcomes including its own crop-growth-model prescription-map technology; the Smart Farm Technology and Systems State Key Lab jointly with HIT and Northeast Agricultural University, co-developing 天工开悟 (units/china-agricultural-llms-landscape.md), the 寒地作物大模型 (cold-region crops model) and the BDH-Rice crop model.

Beidahuang — the group frame and the money (provincial MARA / Heilongjiang Daily, 31 March 2026, read in full)

The group level adds the institutional and economic layers:

• 《集团智慧农业行动方案（2026-2030年）》 issued, with core tasks “build smart farms, create model zones” under a three-tier 集团统筹、产业协同、农场落地 (group coordinates / industries coordinate / farms implement) system — the group’s own five-year smart-agriculture plan, sitting inside a wider new-quality-productive-forces action plan (2024-2030) adopted October 2024 and folded into the group’s 15th FYP strategic development plan in January 2026.
• Infrastructure: 25,300 皮长公里 of five-level fibre network; >90% of farms with 四情监测 (four-situation monitoring) equipment; ~203 TB stored data; 270,000+ plots each with a unique digital identity; 智壤’s holographic one-map covering 21 themes with digital files for 4,800+万亩 of farmland, 600,000 planting households and 760,000+ farm machines.
• The 农服 App is the economic datum: cumulative online service fees of CNY 101.8 billion and 250 million mu-times of machinery dispatch. This is the corpus’s first hard revenue-side figure for Chinese agricultural AI services — and it is unaudited (G-481). The platform’s provincial context: 2,111.4万亩次 of agricultural socialised services delivered.
• The machinery base beneath it: 680,700 units of farm machinery, 14.24M kW total power, mechanisation stable above 99.7% — the physical preconditions the corpus’s catalogue-precondition analysis predicted would gate autonomous agriculture.

Heilongjiang province — the programme frame

Heilongjiang is the first province to issue a smart-agriculture implementation plan (《黑龙江省大力发展智慧农业实施方案》) through the provincial government office, with a route-map and timetable; Beidahuang and Jiamusi were designated the province’s two smart-agriculture model zones in 2025. Provincial work-dynamics give a fleet-level telemetry picture: 16,000 new smart plant-protection drones in a year (provincial fleet ~33,000), 700M mu-times of aerial operations, 121,800 machinery-monitoring terminals, 4.007M mu of drip irrigation — province-reported with no method note (V1-official figures, G-481). The same documents place Beidahuang’s Weishan farm among MARA’s 16 national 2025 smart-agriculture typical cases and the group’s 主粮种植数字管控技术 among the ten nationally promoted technologies. A provincial page (September 2026, discovery pass, not directly read) reports CNY 230 million cumulative invested across the 40 pilot farms. Suihua runs a “1+2+3+10” smart-agriculture system with 32 smart farms and a national BeiDou+ smart-agriculture pilot. The 2026 provincial government work report commits to the “人工智能+” action and a national “AI+” agriculture pilot-testing (中试) base — which HIT’s Liu Jie team is building toward (units/china-agricultural-llms-landscape.md).

The XPCC — division-level only

No corps-level (兵团) AI or smart-agriculture programme document surfaced in this pass — a genuine negative finding (G-479). What exists is division-level and citable but scattered: the 8th Division Shihezi cotton-production LLM on domestic GPU compute with air-ground-space sensing (兵团农业农村局-affiliated media, with a named bureau official quoted — the closest thing to an XPCC AI statement found, but the model has no named developer, name or parameter scale); the first XPCC agricultural data product (新疆耘筹’s corn whole-cycle product registered at the Guangzhou Data Exchange, April 2025, digital services on 4M+ mu, LLM-generated “health-CT” field reports); the 10th Division Beitun “数字十师” app (2025; five middle platforms, 10+ systems); the 6th Division Qitai farm model (+30% drip uniformity, −15% cost). Adjacent but not XPCC: Changji prefecture’s three-year unmanned-machinery-swarm project and Weili’s 3,000-mu two-person super cotton field. Read structurally: the XPCC behaves like 14 Beidahuangs without a group-level platform company — deployment without an integrating estate.

What this unit is doing in the taxonomy

Anchors the state-farm deployment-vehicle cell — the layer where Chinese smart-agriculture policy becomes estate operations. Distinct from:

• units/china-ground-robotics-unmanned-farms.md — the machinery/robotics layer and the catalogue-precondition story; this unit holds the data-platform, action-plan and service-economics layers and deliberately does not restate the robotics detail.
• units/china-agricultural-llms-landscape.md — the model population; Beidahuang appears there as a deployment surface (Tiangong Kaiwu’s 114 farms) and here as an estate.
• units/china-smart-agriculture-action-plan-2024-2028.md — the national plan that names Beidahuang in its addressee list, the corpus’s first hard evidence that state-farm groups are formal implementation vehicles; this unit is what that addressee actually did.
• units/china-no1-central-document-2026-ai.md — the political-mandate and 15th FYP frame above both.

New gaps surfaced by this unit

• G-479 (new): XPCC corps-level smart-agriculture programme — nothing found; only division-level cases. Track: any 兵团-level plan document, the Shihezi cotton LLM’s developer and parameters, corps-level data-platform procurement.
• G-480 (new): the text of 《黑龙江省大力发展智慧农业实施方案》 — the plan is described in provincial work-dynamics but its full text was not located this pass; treat the route-map claims as cited-at-second-hand until read.
• G-481 (new): independent verification of the Beidahuang estate figures — the CNY 101.8bn 农服 App fee figure, the 48-satellite monitoring claims and the pilot-farm outcome percentages are company/provincial self-reports with no audit; no regulated disclosure located.

New contested claims surfaced

• C-361 (new): “State farms are the deployment vehicle that makes Chinese agricultural AI real at scale — the estate layer, not the farmer layer.” Supported by the estate’s documented scale (48M+ mu digital files, 111 farms, 99 algorithms in production) and its institutional completeness (own information company, own lab, own service platform); countered by the fact that every operating figure is self-reported, the service-fee total has no independent check, and the model layer it hosts is itself unaudited (G-476).