Produced by | He Xi

Layout by | Ye Yuan

In 2026, financing in the embodied AI sector continued to surge: Galaxy General secured RMB 2.5 billion, Qianxun Intelligence raised RMB 3 billion in 30 days, and ISpace closed a Series B round exceeding RMB 1 billion. In Q1 alone, industry financing approached RMB 20 billion.

But financing amounts have never been the industry's thermometer—orders are.

ISpace secured nearly RMB 500 million in orders, UBTECH's total order value surpassed RMB 1.4 billion, and State Grid invested RMB 6.8 billion to procure 8,500 devices. These figures reveal more than any financing round: Who is truly paying for robots? Which scenarios have achieved closed-loop commercialization?

The author previously dissected two facets of embodied AI: (1) how upstream core component companies quietly generate profits (Blue Point Touch, Kunwei Tech, and Quan Zhibo now supply in bulk) and (2) the fusion battle between VLA large models and 'brain' technical routes. However, the final link in the industrial chain—downstream actual buyers—has lacked systematic empirical reporting.

This article systematically reviews publicly disclosed orders from the past year to create the first 'Order Truth Map' of embodied AI commercialization, answering three progressive questions: (1) Why does willingness to pay vary drastically across scenarios? (2) Which companies have secured tangible purchase contracts? (3) What competitive landscape do order distributions reveal?

01 Scenario Pyramid: How Willingness to Pay Stratifies from Top to Bottom

Disclosed order distributions reveal a clear 'three-tier pyramid' structure for embodied AI commercialization. The higher the tier, the stronger corporate payment willingness and larger order scales become.

Pyramid Apex: Industrial Manufacturing—The Most Definitive Battleground

Industrial manufacturing represents the most certain primary market for embodied AI adoption. Sectors like automotive final assembly, 3C electronics, semiconductors, and precision machining face dual pressures of 'recruitment difficulties' and yield improvement, with machine substitution ROI precisely calculable. MIIT and SASAC jointly promoted the formation of 10,000-unit deployment capacity by end-2026. IDC predicts global humanoid robot shipments will exceed 50,000 units in 2026, up 178% YoY.

UBTECH delivered over 1,000 units in 2025 (80%+ for industrial scenarios), with annual orders surpassing RMB 1.4 billion. Its 2026 production capacity will reach 10,000 units. Mainstream automakers including BYD, Geely, FAW-Volkswagen, and Audi FAW have procured its Walker S series for training, with cumulative intention (tentative) orders exceeding 500 units. ISpace signed a 3-year, 1,000-unit contract with semiconductor display giant HKC (valued at nearly RMB 500 million), covering warehousing, component assembly, and quality testing—recognized by Morgan Stanley as the 'world's largest single order for productivity-focused robots.' Qianxun Intelligence's 'Xiaomo' deployed en masse at CATL's Zhongzhou base for battery PACK line high-voltage testing, marking the world's first battery production line with scaling (scaled) humanoid robot adoption. Galaxy General signed a strategic partnership with Baida Precision for over 1,000 embodied AI robots, fully opening its high-precision component production lines for validation—signaling the official start of scaled applications in precision industries.

Pyramid Body: Public Utilities & Energy Infrastructure—New Growth Pole Driven by Central SOE Group Purchasing

State Grid's 2026 special plan earmarked RMB 6.8 billion to procure ~8,500 embodied AI devices across power inspection, live-line operations, emergency rescue, and warehousing logistics, including 5,000 quadruped inspection robot dogs. This marks China's largest single embodied AI group purchase, with central SOE budget endorsements signaling the sector's transition from 'corporate pilots' to 'industry standards.'

Pyramid Base: Commercial Services & Household—Pilot-Centric, No Scaled Adoption Yet

Scenarios like hotel delivery, retail guidance, and restaurant services have clear market demand but longer ROI cycles, currently relying on 'brand tech appeal'-driven pilot purchases. Galaxy General achieved normalized operations across nearly 100 retail stores with over 300,000 cumulative orders but lacks scaled procurement. Household scenarios remain virtually devoid of verifiable scaled orders—the journey from lab to living room proves far longer than from production line to factory.

02 Order Truth: Who Secured Tangible Purchase Contracts?

Financing can rely on PPTs, but orders demand delivery. Systematic review of past year's disclosed orders clarifies embodied AI's 'order truth.' Note that total orders remain 'pilot-dominated, batch-auxiliary,' with 10,000-unit shipments not yet industry norm, though growth signals are clear.

Representative Order Overview

Differentiated Order Paths of Leading Enterprises

'Brain-Centric' and 'Body-Centric' robot leaders have formed clear commercialization paths. UBTECH built the most complete automotive manufacturing client cluster, with humanoid robot orders exceeding RMB 1.4 billion—the industry's densest commercialization data. Its 'industrial + government + scenario' multi-dimensional order structure achieves scale effects.

ISpace's self-developed AlphaBrain embodied large model achieves 20ms collision reflection latency—a core safety metric enabling deep deployment from wafer handling precision operations to HKC's full-process factories. HKC's 1,000-unit order proves deep industrial commercialization feasibility.

Qianxun Intelligence applies its general large model and data collection iteration capabilities to CATL and JD Retail scenarios, establishing benchmark cases at both industrial manufacturing and commercial service ends while accumulating data barriers via 'VLA + data flywheel' paths.

Zhiyuan Robotics deep cultivation (deeply cultivates) 3C manufacturing, with its Elf G2 achieving 140-hour continuous operation at Longcheer Technology's production line and a 12.97-second work cycle at Joyson Electronics—verifying product engineering capabilities. Its 10,000th general-purpose embodied robot rolled off the line, setting a humanoid robot mass production speed record.

Galaxy General secured 1,000-unit orders in precision manufacturing, gaining strong first-mover advantages across policy cycles and capital rhythms.

Key Signal: Upstream Components Scale Ahead of Complete Machines

While complete machine companies rely on financing narratives, upstream suppliers already profit through orders. Blue Point Touch shipped over 100,000 joint force sensors in 2025 (expected 400,000 in 2026), achieving three consecutive years of doubled revenue and profitability while raising market share from 62% to 72.6%. Kunwei Tech holds over 80% market share in collaborative robot segments, supplying six-axis force sensors in bulk to UBTECH, Galaxy General, Zhiyuan, and Xinghai Tu, with 500 units delivered in three weeks. Quan Zhibo's 2026 self-built integrated joint automation line achieved full production, reaching million-unit annual capacity—far exceeding the original 2027 target.

Financing can inflate valuations, but only orders sustain profits. While complete machine makers climb toward 10,000-unit production, upstream suppliers have already secured real purchase orders—a critical signal of the industry's shift from 'concept hype' to 'supply chain substance.'

03 Buyer Logic & Competitive Landscape: Who Defines 'Payment Standards'?

Orders reflect buyers' true decision-making logic. Understanding why manufacturers pay reveals competitive landscape evolution.

Three Core Buyer Pain Points

(1) Reducing labor costs amid 'recruitment difficulties.' Manufacturing faces a general reluctance among young workers to join factories. Zhiyuan's Elf G2 achieved 140-hour continuous operation at Longcheer with 99.5% success, replacing two workers' workload and completing debugging to integration in four months. UBTECH's Walker S series training tasks at multiple automakers target repetitive physical labor replacement.

(2) Improving yield and precision consistency. Electronics and precision parts production have extremely low error tolerance, with human operations prone to fatigue-induced defect rates. ISpace's AlphaBot 2 at HKC achieved 4,000 daily moves, reducing fragmentation rates by 90%. Qianxun Intelligence's 'Xiaomo' eliminated safety risks from high-voltage arcing during hundreds-of-volts plugging operations at CATL—a risk humans cannot fully mitigate. Yield improvements directly convert to financial returns, with payback periods under 12 months in high-value-added scenarios.

(3) Forced capacity expansion and flexible manufacturing. Some firms, facing sustained order growth, view automation as a rigid means to expand capacity. ISpace covered material handling, labeling, and storage at Dongfeng Liuzhou Automobile, achieving the first large-scale application of domestic embodied large models in automotive manufacturing. Qianxun Intelligence partnered with global precision parts leader Schaeffler to deploy production line closures in box folding and packaging, constructing a complete 'real data-embodied model-real scenario' data flywheel.

Industry Variations in ROI Calculations

Investment payback periods vary significantly by scenario: semiconductor/automotive high-value-added scenarios achieve payback in <12 months; 3C manufacturing/general assembly in 1.5–2 years; commercial services/retail exceed 3 years, currently brand-driven.

Accelerating Competitive Differentiation

UBTECH built the most complete automotive manufacturing client cluster, with orders exceeding RMB 1.4 billion and 10,000-unit production capacity—the industry's densest commercialization data.

ISpace achieved the deepest vertical penetration in semiconductor panel manufacturing, with HKC's 1,000-unit order proving industrial deep commercialization feasibility.

Galaxy General secured 1,000-unit orders in precision manufacturing, gaining clear first-mover advantages.

Zhiyuan Robotics accelerates 3C manufacturing layout (layout), with 10,000 units rolled out setting mass production speed records.

Qianxun Intelligence establishes dual industrial + retail benchmarks, continuously accumulating data barriers via flywheels.

Upstream component firms achieve profitability + bulk supply, with order certainty exceeding some complete machine makers.

Policy Catalyst: From 'Corporate Pilots' to 'Industry Standards'

State Grid's RMB 6.8 billion group purchase and MIIT-SASAC's joint 10,000-unit deployment push—central SOE procurement is unlocking a market vaster than manufacturing. The 2026 Humanoid Robot and Embodied AI Real-Scenario Training Initiative targets 10,000-unit deployment capacity by year-end. Policy catalysts plus accelerating industrial orders are transforming embodied AI commercialization from a 'few players' game' into an 'industry-wide competition.'

But true winners will emerge from order delivery battles—only firms converting 'tentative orders' into 'batch deliveries' and 'pilot validations' into 'standard configurations' can define next-stage market rules.

Epilogue

Returning to the opening question: Beyond financing amounts, what is the 'order truth' of embodied AI?

The answer is clear—industrial manufacturing has emerged as the definitive battleground, with leading firms validating commercial models from 0 to 1. Central SOE group purchasing is unlocking a second growth curve, while upstream components have already scaled and achieved profitability. However, 10,000-unit shipments remain non-standard, commercial services await ROI inflection points, and hype still intertwines with reality.

Financing can inflate valuations, but only orders sustain profits. 2026 marks the critical first year of scaled embodied AI deployment, with competition shifting from 'prototype demonstrations' to 'real order' hard battles. Firms crossing the 10,000-unit threshold first will secure comprehensive first-mover advantages in mass production efficiency, supply chain bargaining power, and product iteration.

True winners will emerge from order delivery battles.