JD.com Q2 2026 Earnings: 80 RoboBase Robotics Hubs, 60 Smart Warehouses, and 10-Second Dual-Arm Picking Signal a Real-World AI Infrastructure Push

Forum topic · 小凯 · 2026-08-13

Summary

JD.com's Q2 2026 results reveal a strategic pivot toward embodied AI infrastructure. Revenue reached 346.4 billion yuan (-2.9% YoY), net profit rose 14.5% to 7.1 billion yuan, service revenue climbed to 22.9%, and R&D spending accelerated 53.2%. JD plans 80+ RoboBase robotics hubs over 5 years, with the first under construction in Guangzhou offering data collection, testing, and maintenance. The company launched the open-source EgoLive first-person dataset and three physics-world models (JoyAI-Echo, JoyAI-VL-Interaction, JoyAI-Video-Edit). Deployed assets include the ZhiLang goods-to-person system in 60+ global warehouses, a 10-second dual-arm YiLang picker, thousands of autonomous delivery vehicles across 20+ Chinese provinces, and JoyAvatar digital humans serving 80,000+ merchants. Unlike Tesla, Figure AI, or Unitree, JD treats robots as shared infrastructure rather than products. Management cautions current metrics do not yet quantify AI-driven revenue or cost gains, making the Q3 earnings the true validation checkpoint.

Key points

Financial backdrop

Q2 revenue: 346.4 billion yuan, down 2.9% year over year
Q2 net profit attributable to shareholders: 7.1 billion yuan, up 14.5%
Service revenue: 79.3 billion yuan, now 22.9% of total
Half-year R&D spending accelerated 53.2%, the third consecutive quarter of acceleration
JD Logistics revenue +26.5%, JD Health +17.2%, JD Industrials +16.3%
Product GMV is falling while R&D and service revenue rise, suggesting JD is trading near-term profit for the next technology cycle

Three infrastructure commitments

1. RoboBase robotics hubs — 80+ over 5 years The first RoboBase project broke ground in Guangzhou during Q2, offering robot data collection, testing, scenario application, and maintenance. JD intends to build 80+ such hubs nationwide to create a closed "R&D–manufacturing–application–service" loop for physical AI.

2. EgoLive first-person dataset JD opened China's first embodied-data collection community in Suqian, Jiangsu, and open-sourced EgoLive, described as the industry's largest first-person human dataset. Data comes from real retail, logistics, health, and industrial workflows, with 100+ universities and research institutes applying for access.

3. Three open-source models

JoyAI-Echo: long-form audio-video generation framework
JoyAI-VL-Interaction: real-time video vision-language interaction model
JoyAI-Video-Edit: real-time streaming video editing model

Three deployment milestones

Warehousing

ZhiLang goods-to-person solution live in 60+ warehouses worldwide, including the UK and Germany
YiLang robotic arm upgraded from single to dual-arm, completing recognition, grasping, and stacking within 10 seconds

Last-mile delivery

Thousands of autonomous delivery vehicles operating across 20+ Chinese provinces; Shenzhen launched the first nighttime unmanned delivery route
Drone network in Sichuan's Zizhong county covers 78 administrative villages, with the fastest mountain crossing in 7 minutes

Merchant services

JD digital human JoyAvatar has served more than 80,000 merchants
Q2 livestream accounts using digital humans tripled YoY; daily active accounts grew 6x
During 618, both livestream hours and GMV from digital humans more than quadrupled YoY

Why JD's path differs

Tesla Optimus pursues full-stack: in-house robot + in-house data + in-house scenarios
Figure AI pursues single-point breakthroughs: general humanoid robot + BMW factory deployment
Unitree pursues scale: cost-effective hardware + education/inspection scenarios
JD instead treats robots as shared infrastructure: 80+ RoboBase hubs combine "4S dealership + repair station + data center + training ground," opening every JD scenario as a real-world training environment

The open question

Management explicitly admitted on the earnings call that current figures do not yet quantify how much AI investment has lifted revenue, profit, or reduced costs. The next validation checkpoints are: 1. Whether ad conversion improves as digital-human livestreaming scales 2. Whether per-order fulfillment cost is actually lowered by dual-arm YiLang robots 3. Whether external technical services revenue reaches a quantifiable share of total revenue

If two of these three turn positive within six months, the 5-year RoboBase pledge shifts from "laying foundations" to "paving runways." The Q3 earnings release will be the real proof point.