On July 17, 2026, at the World Artificial Intelligence Conference (WAIC) in Shanghai, the humanoid robot "Spirit G2 Max," independently developed by AgiBot, drew widespread attention. As a safety-grade heavy-duty embodied robot, it is slated to be deployed within JD Logistics' self-developed "Smart Wolf" goods-to-person warehousing system, operating around the clock to handle material transport and palletizing tasks. This system marks the first real-world application of embodied robots within the logistics industry, bridging the gap between exhibition displays and fully operational warehouses.
Just over three weeks later, on August 13, JD.com (HKEX: 9618; NASDAQ: JD) released its second-quarter and interim financial results, revealing second-quarter revenue of 346.4 billion yuan—a year-on-year decline of 2.9%—while operating profit swung from a loss of 900 million yuan in the prior year to a profit of 4.5 billion yuan, with operating margin climbing from -0.2% to 1.3%. For the first half of 2026, total revenue reached 662.1 billion yuan, with service revenue surging 12.9% year-on-year, showcasing strong growth across various service segments. Tucked within the same report was a seemingly contradictory detail: R&D investment grew 53.2% year-on-year, accelerating for three consecutive quarters to reach 14.165 billion yuan, nearly matching the full-year 2024 expenditure of approximately 17 billion yuan, with a substantial portion channeled into AI initiatives.
According to JD.com's roadmap, dedicated AI R&D spending across the entire ecosystem is set to increase by over 200% in 2026, with resources concentrated on embodied intelligence, large-model computing power, and industrial scenario deployment. Warehouse automation projects—including the Smart Wolf storage system and Spirit G2 Max humanoid robots—serve as the core carriers for AI's physical implementation.
Why Just One Financial Metric?
While profits have turned positive, R&D spending has doubled—JD.com's central question is remarkably straightforward: Can AI actually generate profits, and if so, how? Rather than selling AI directly, JD.com embeds AI within the cost structure, operational efficiency, and experience thresholds of every single order. There is no standalone "AI business" revenue line item—AI is woven into the profit improvements of each transaction, the efficient applications across the industrial chain, and the daily lives of consumers.
Notably, as algorithms converge and open-source model capabilities narrow, the true competitive moat is shifting from "who has the stronger model" to "who possesses the scenarios and data where AI can genuinely perform real work." In this direction, JD.com is quietly positioning itself as the "shovel seller" of the trillion-yuan embodied intelligence industry—amassing vast amounts of embodied data while retraining its blue-collar workforce as robot operators. Leveraging over two decades of industrial depth, JD.com is feeding AI with real-world operational data, using AI to reconstruct supply chain efficiency, and reinvesting those efficiency gains back into AI advancement. The ultimate objective is clear: to build a world-class JoyAI foundational model matrix, propel AI from the digital realm into the physical world, and establish the largest global physical-world operations center grounded in JD.com's authentic industrial scenarios, empowering industries and households at scale.
Earnings Contrast: How AI Enhances the Profitability Statement
To comprehend JD.com's AI strategy, one must begin with this latest "unusual" financial report. In Q2 2026, revenue stood at 346.4 billion yuan, down 2.9% from the same period in 2025, which JD.com attributed to "a high comparison base from 2025 and weak consumer electronics demand." Despite revenue pressure, a clear inflection point emerged on the profit side: operating profit reached 4.5 billion yuan versus a loss of 900 million yuan a year earlier; non-GAAP operating profit hit 5.5 billion yuan compared to just 900 million yuan previously; and net profit attributable to ordinary shareholders rose approximately 15% to 7.1 billion yuan.
Even more striking was the performance of JD Retail, the core segment: Q2 operating profit hit 13.5 billion yuan with an operating margin of 4.6%, a historic high for the 618 Grand Promotion season. CEO Sandy Xu attributed the improvement to "solid profitability in the core JD Retail business and continued losses narrowing in JD's food delivery segment." This suggests that despite new ventures (food delivery, Joybuy, Jingxi) and the AI innovation segment remaining in strategic investment phases, the operational efficiency gains from the core business have been sufficient to offset and even exceed these losses.
Beneath this performance improvement lies a hidden message: JD.com's AI investment has not weighed on profits—it has actually become a lever for profit enhancement. How is this possible? The answer resides in "cost structure." JD.com's business model fundamentally revolves around delivering goods from production to consumption with maximum efficiency across retail, logistics, industrial, and health operations. AI plays an indispensable role in the retail fundamentals of "experience, cost, and efficiency": when the Super Brain large model calculates optimal routes for billions of packages in three seconds, when automated warehouses push sorting costs lower, and when industrial agents compress data governance from months to hours, every yuan saved in fulfillment costs manifests directly on the income statement.
A tangible example is the Smart Wolf warehouse, where high automation continuously reduces per-unit turnover and fulfillment costs, achieving payback in under three years. At the Smart Wolf facility within Beijing's Asia No.1 Daxing Smart Industrial Park, a product's location is locked within five seconds of a consumer placing an order; climbing robots retrieve bins along racks while ground transport robots relay them to picking staff, completing the process in under one minute. Outbound efficiency has tripled compared to manual operations, with a picking accuracy rate of 99.99%. In Milton Keynes, UK, JD Logistics operates an automated warehouse spanning over 3,000 square meters with nearly 200 Smart Wolf units, achieving roughly a fourfold improvement in picking and outbound efficiency. As per-package handling, sorting, and delivery costs continue to decline alongside rising high-margin revenue streams like customer acquisition and advertising, profit margins improve with AI's assistance. This underscores that JD.com is not crafting a new AI narrative but rather using AI to make existing operations cheaper, faster, and better.
Importantly, JD.com's AI investment follows an "endogenous" self-developed route rather than an outsourced model. The company has fully self-developed its JoyAI foundational large-model matrix and built its own JoyScale computing platform, maintaining autonomous control over computing adaptation and scheduling—a core long-term cost. This contrasts sharply with enterprises that purchase third-party public cloud services and pay per inference: endogenous development entails higher upfront R&D and hardware costs, but as business scale expands, marginal computing costs continuously decline; the pure outsourcing model is asset-light initially but requires sustained external infrastructure payments, lacking control over underlying computing costs.
JD.com began its technology transformation as early as November 2016, when founder Richard Liu announced plans for the company's second twelve-year phase, committing to full technological advancement with heavy investment in AI and robotics automation. Financial reports show annual R&D spending at the hundred-billion-yuan level from 2018 to 2024, reaching 22.2 billion yuan in 2025 and 14.165 billion yuan in the first half of 2026 alone. At the 2026 618 launch event, Cao Peng, Chairman of JD.com's Technology Committee and President of JD Cloud, revealed that AI-related R&D investment across the ecosystem would grow by over 200% this year, placing JD.com among the industry's top tier.
While there were prior doubts about why a retail-focused enterprise would heavily invest in long-cycle technologies like automation and AI, this long-term investment logic is now delivering tangible returns. However, it's crucial to acknowledge the other side: a substantial portion of Q2 profit improvement stems from delivery loss reduction, category mix optimization, and high-margin revenue growth such as advertising. AI's contribution to profits is structural and gradual. When isolated, JD.com has no "AI business" revenue line—but it is precisely this invisibility that defines JD.com's AI: embedding AI within the cost improvements of every order. During the H1 2026 earnings call, CFO Ian Shan projected continued supply chain efficiency gains driving retail gross margin improvements, while noting that AI-related R&D expenses would keep growing: "We believe these investments will gradually translate into operational dividends. Long-term, we remain confident in achieving a high-single-digit profit margin target."
Industrial Physical AI: Unlocking Efficiency and Cost Reduction
Beyond the balance sheet numbers, AI's tangible impact on JD.com's operations—particularly on logistics and supply chain efficiency—represents the core of its AI deployment. In 2025, JD Logistics completed a full operational loop of "cloud intelligence to terminal execution" through the comprehensive upgrade of its Super Brain Large Model 2.0 and the large-scale deployment of the "Wolf Robot Corps," signaling the accelerated formation of a self-evolving JD super supply chain.
The Super Brain large model serves as the central nervous system of JD's super supply chain. In 2023, JD Logistics pioneered the industry's first large-model-powered intelligent logistics brain, "Super Brain." After two years of iteration, on September 25, 2025, Super Brain 2.0 was unveiled, fully embracing Agentic intelligence to enable autonomous decision-making for smart devices. Its core capabilities lie in extreme efficiency and dynamic optimization: based on digital twin intelligent decision systems, it compresses simulation time for billion-scale packages to under three minutes, and combined with model acceleration technology for tens of millions of variables, it dynamically adjusts optimal logistics planning. The model is now applied across over a hundred scenarios, demonstrating formidable practical value.
During the 2026 618 period, Super Brain 2.0 was deployed at scale, achieving deep AI application across over 1,000 core logistics and supply chain scenarios. Acting like "real-time navigation," it dynamically plans optimal transport routes for parcels, directly reducing empty vehicle mileage and transfer costs. At the terminal execution level, JD Logistics deployed the "Wolf Robot Corps"—a family of terminal execution equipment named for wolves' efficient pack coordination—incubating and scaling over ten robot types, including Smart Wolf, Ground Wolf, Sky Wolf, Lone Wolf, Flying Wolf, and automated sorting walls.
The Smart Wolf, an industry-first goods-to-person system, has been implemented in over 60 warehouses globally as of Q2, including locations in the UK and Germany. The "Strange Wolf" features an "industrial-grade dexterous hand" that completes a full sequence—from recognition to suction, grasping, and stacking—in just 10 seconds, making it the only embodied intelligence integrated into real industrial practice. The Wolf Robot Corps has been deployed across logistics supply chains in over 10 countries and more than 20 Chinese provinces, achieving full-chain scale coverage. In last-mile delivery, over a thousand autonomous vehicles operate routinely across 20+ provinces, with JD.com launching nighttime delivery routes in Shenzhen. In Zizhong, Sichuan, JD has established one of China's most extensive drone-to-village delivery networks, crossing mountainous terrain in as little as seven minutes and covering 78 administrative villages—together forming the "capillaries" of JD's logistics system to resolve last-mile challenges.
From Super Brain 2.0 to the Wolf Robot Corps, physical AI now spans the entire logistics operational chain—storage, picking, transport, sorting, and last-mile delivery—representing a large-scale, genuine industrial practice of physical AI in the supply chain domain. JD Logistics plans to invest in 3 million robots, 1 million autonomous vehicles, and 100,000 drones over five years to completely reconstruct its supply chain operations. In the future, robots will handle everything from procurement and warehousing to transport and home delivery, embodying JD's vision for a physical-world operations center.
The "Super Brain + Wolf Corps" synergy extends beyond internal efficiency gains; these capabilities are also commercialized through a platform for industry clients, providing solutions to retail, home appliances, finance, insurance, and power sectors, achieving cost reduction and efficiency gains for over a hundred brands. A prominent domestic book publisher serves as a case study: within its Lingnan warehouse, JD Logistics deployed the Smart Wolf shuttle version for eightfold high-density storage and efficient in/outbound operations, paired with automated sorting walls to resolve forward and reverse sorting challenges, lifting overall warehousing efficiency by over 1.5 times, reducing operational space by more than 50%, and maintaining a sorting accuracy of 99.99%. This isn't simply "moving goods online"—it replicates the efficiency of JD's own warehouses within clients' supply chains.
These examples reflect JD Logistics' objective: building a Chinese super supply chain to reduce society-wide logistics costs and address chronic industry inefficiencies. This illustrates AI's "efficiency enhancement and cost reduction" logic at JD's industrial end—while not directly generating new revenue, it transforms wasted fulfillment costs in the real economy back into profit. A securities analyst noted that JD.com isn't running benchmark tests in a laboratory; it's moving, sorting, and delivering daily under real operating conditions, using genuine logistics data to refine models. This is precisely why JD.com can genuinely implement physical AI in logistics and supply chains, creating an AI-era super supply chain.
Consumer AI Suite: Elevating Experience to Unlock New Markets
While AI drives efficiency and cost reduction on the industrial side, it generates revenue growth and enhanced experiences on the consumer side—the other key pillar of JD's AI strategy, distinct from industrial applications. On the consumer front, AI transforms from a cold "tool" into a warm assistant, encompassing "Jingyan," an intelligent shopping decision assistant; JoyAvatar, a virtual livestream digital human; and JoyInside, an embodied intelligence platform offering a wealth of AI-powered products.
Most directly consumer-facing is Jingyan, JD's self-developed AI agent that condenses JD's understanding of products and supply chains into a conversational shopping gateway, enabling more precise matching of products and services to consumer needs for a more efficient and convenient shopping experience. In Q1, Jingyan's quarterly active users approached 80 million, up over 200% year-on-year.
The digital human JoyAvatar has captured broader attention, serving over 80,000 merchants cumulatively. In Q2, the number of accounts using digital humans for livestreaming tripled year-on-year, with daily active accounts surging sixfold. During 618, digital human streaming duration and transaction volume both grew over fourfold year-on-year, substantially reducing merchants' livestream labor costs while driving GMV growth several-fold through multi-brand IP collaborative streaming.
JoyInside, however, represents the most imaginative consumer-side AI application. As JD's hardware AI empowerment open platform built on the self-developed JoyAI large model, it implants "AI brains" into physical hardware—home appliances, toys, lighting—endowing them with conversational ability, proactive needs understanding, long-term memory, and emotional interaction. This serves as JD's core vehicle for embodied intelligence and a physical-world AI entry point. Through JoyInside, JD aims to make warm terminals ubiquitous.
Recognizing the growing demand for proactive companionship in the maternal and companion economy, the JoyInside team, in collaboration with industry associations, identified the AI toy sector's explosive potential: the domestic market has already reached the tens of billions scale, with medium-to-long-term potential entering the hundreds of billions, growing at a compound annual rate far exceeding traditional toys. By integrating JoyInside with JD's in-house brand Jingzao supply chain advantages, JD launched an AI plush toy with high emotional interaction capability in under a month. JoyInside played a pivotal role by providing foundational model capabilities, enabling hardware intelligence and tailoring base capabilities to device constraints for optimal interaction experiences.
JoyInside's distinctiveness lies in its customized interaction design across age groups: child-oriented AI language is lively and age-appropriate; adult responses are more dynamic and energetic; and the "Chattering Parrot" product for seniors draws inspiration from bird-walking scenes, with model design prioritizing stability, gentleness, and caution to address complex emotional needs. Long-term memory functions allow the toy to remember children's preferences, building lasting emotional connections. Within six months of launch, the product climbed to the top of JD's plush toy category satisfaction rankings. One user shared a personal account: their child previously resisted waking up, but with the JD AI companion, the child now rises proactively, treats the toy as a role model, and even named it—a form of proactive psychological guidance that traditional passive companionship cannot achieve.
To date, JoyInside has partnered with nearly 200 hardware brands, providing intelligent interaction capabilities across categories including AI toys, robots, home furnishings, home appliances, and medical devices. Examples include Xinfei AI voice-controlled air conditioners enabling elderly users to simply say "set temperature to 26 degrees," and Bangbang's "Guardian Star" wheelchair that engages seniors in multi-turn natural conversation, plays traditional opera, and allows children to monitor location in real-time and set "safe care fences" via the JD Health app. Yuanluobo's AI chess robot, once integrated, offers not just gameplay but chess instruction, AI education, and companionship. These traditional physical devices have gained "souls that chat and express emotions" through JoyInside, reshaping user experiences and unlocking larger consumer markets.
A JoyInside lead revealed that approximately 50% of JD's AI-category product market share comes through JD's channels: "If JoyInside helps diverse products achieve AI transformation, AI product mindshare will stay with JD." In contrast to the industrial end's extreme efficiency and cost reduction, JD's consumer-side AI core value centers on "experience" and "revenue growth": Jingyan improves product discovery efficiency, digital humans lower merchant operational barriers, and JoyInside expands AI product supply. Through AI, JD is shifting consumer demand from zero-sum competition to incremental blue oceans. Efficiency gains and revenue growth act as two fulcrums of the profit statement: the industrial side compresses costs while the consumer side elevates value, jointly supporting the profit inflection point visible in JD's financials.
Embodied Intelligence: Positioning for Data and Talent
While JD has focused on using AI to enhance efficiency and revenue across industrial and consumer segments, its strategy in the trillion-yuan embodied intelligence arena centers on becoming a "shovel seller." These investments are building a formidable moat for JD's future profitability. A senior executive at a leading robotics company observed: "In the AI industry, shovel sellers profit first. Data is the essential requirement for embodied intelligence—regardless of which robot or model ultimately prevails, they all depend on data supply."
The industry's most pressing challenge is that robots have well-developed "cerebellums"—capable of dancing and boxing—while their embodied models, the "brains," lag behind. The root cause is an acute scarcity of real-world scenario data. According to NetEase Technology reports, by early 2026, globally available high-quality real-world physical interaction data totaled only approximately 500,000 hours, while industry consensus indicates that training general-purpose embodied models requires at least ten million hours—a shortfall exceeding 99%. Shenwan Hongyuan's June research report declared 2026 the pivotal year for physical AI, describing data as the core resource equivalent to lithium ore in the embodied intelligence era. The industry faces a severe physical AI data bottleneck, with VLA models (vision-language-action) requiring trillion-scale physical interaction data versus the existing million-scale public datasets—an enormous gap.
A fund manager focused on robotics industry investment candidly acknowledged: "The biggest challenge for humanoid robots today is large models, and the largest shortcoming of those models is the shortage of physical interaction data." JD.com identified this industry pain point early and has launched an unprecedented embodied intelligence data collection initiative: leveraging its super supply chain and massive real business scenarios, JD plans to build the world's largest and most comprehensive embodied intelligence data collection center, mobilizing 600,000 people for a historic data collection effort targeting 10 million hours of real-world scenario video data within two years, alongside 1 million hours of robot-specific data. The collection center covers five core scenarios—logistics warehousing, industrial manufacturing, healthcare, home services, and urban operations—employing self-developed wearable devices for first-person egocentric collection, supplemented by teleoperation. The nation's first embodied data collection community has been established in Suqian. To date, JD has completed 2 million hours of data collection, making it the industry's largest dataset. High-quality real-world data is fundamental for transitioning physical AI from laboratories to genuine industry—every frame from retail, logistics, health, and industrial scenarios corresponds to authentic physical operations and industrial requirements.
JD has open-sourced EgoLive, the industry's largest first-person view dataset, and its data trading platform has made initial high-precision datasets available to academia, third-party developers, and ecosystem enterprises. More infrastructurally significant is the RoboBase project. In July, JD's first RoboBase site broke ground in Guangzhou, building robot lifecycle industrial infrastructure atop JD's business ecosystem. JD plans to establish over 80 RoboBase robotics bases nationwide within five years, connecting data collection, model training, industrial validation, and after-sales services to create a complete "R&D-manufacturing-application-service" ecosystem loop for the physical AI era.
"JD doesn't require doing all industry data ourselves—we're committed to handling the heaviest part 'for the industry,' leaving the rest to ecosystem partners," said Gong Yicheng, JD Group Vice President and Head of JD Cloud's foundational cloud business, in July. He emphasized that the true industry pain point isn't "lacking data" but "lacking truly valuable, high-quality data."
Closely tied to the "shovel seller" logic is JD's redefinition of employment. At the 2026 APEC CEO Summit China Forum, founder and Chairman Richard Liu stated: "In the future, robots will handle deliveries, eliminating the need for couriers, but I don't want our 700,000 brothers to lose their livelihoods." JD subsequently unveiled its internal "Nirvana Plan": partnerships with 124 schools nationwide to progressively send 700,000 blue-collar workers, including couriers, back to classrooms for robot maintenance and repair training. "When machines malfunction, people must handle it—mechanical things always break down, and breakdowns still require human service," Liu remarked. Through targeted training, JD aims to achieve the "white-collarization" of blue-collar workers, transitioning couriers from physically demanding outdoor work to office-based roles.
This is not merely aspirational. JD has launched one of China's most extensive blue-collar training initiatives, covering 183 frontline service positions including home cleaners, maintenance engineers, auto technicians, and customer service representatives, having trained and supplied over 100,000 professional domestic service personnel. JD Auto Care partners with 110 colleges nationwide, aiming to train 100,000 engineers over five years to serve robot and smart home after-sales maintenance. Behind this "white-collarization" lies a severely underestimated new industry: the robot after-sales market. As millions of robots enter logistics, industry, and homes, maintenance, repair, and operations will generate massive employment—potentially rivaling today's automotive aftermarket. Robots may replace couriers' physical labor, but they create new roles for "robot service providers." JD has launched JoyRobocare, a specialized robot repair service providing professional maintenance for various robot types, now covering major cities in the UK, Germany, France, and the Netherlands. The service currently focuses on embodied and quadruped robots, offering full lifecycle delivery support including doorstep delivery, on-site commissioning, system configuration, and operational guidance.
All evidence points to a pivotal shift: in an era of algorithmic convergence and diminishing open-source capability gaps, whoever possesses the scenarios and data enabling AI to perform real work secures the next-phase entry ticket. JD.com is betting on precisely that ticket. In the robotics gold rush, JD may not produce the flashiest machine, but it is steadily becoming the reliable "shovel seller"—collecting data through its super supply chain and real business scenarios, converting datasets into "brain nourishment" for robots, and transforming 700,000 blue-collar workers into the robot era's "operators." This is the most tangible element of JD's new narrative: AI doesn't reside in some future "to-be-profitable" story—it lives in the profit improvements of every order, in the efficient applications across the industrial chain, in consumers' daily lives, in the trillion-yuan embodied intelligence market, and in every blue-collar worker returning to the classroom.
Comments