How JD.com Built a Hybrid Cloud to Power Its Logistics Empire

01 JD Cloud Evolution

In February 2017, JD.com CEO Liu Qiangdong announced a 12‑year technology transformation strategy, prioritizing cloud computing, big data, AI, and genomics. Since then, JD has invested nearly 75 billion CNY in technology, with JD Cloud now serving over 1,500 large enterprises and 1.52 million SMEs.

Early cloud efforts focused on supporting massive sales spikes, evolving from a centralized architecture to distributed, virtualized, and eventually micro‑service architectures around 2012. By 2014, JD adopted Docker and OpenStack‑nova‑docker, creating its first container platform JDOS 1.0, which powered critical applications like flash sales and order management.

In 2016, JD integrated Kubernetes, storage, and networking into JDOS 2.0, migrating databases from Oracle/SQL Server to MySQL and launching cloud database services.

By 2017, JD leveraged Kubernetes to rebuild its tech stack, deploying databases and big‑data workloads on the platform and introducing the “Archimedes” scheduler, an early hybrid‑cloud unified scheduling system.

02 JD Logistics Cloud Migration

Logistics, a core JD asset, grew from 2,000 employees and 4 billion CNY revenue to 370,000 staff and over 700 billion CNY, driving massive infrastructure demand. Starting in 2015, JD separated its technology division, granting it independence to develop cloud, big‑data, and AI capabilities.

Logistics faced unique challenges: thousands of local data centers, diverse VPC subnets, and heavy reliance on stateful services (databases, Redis, Elasticsearch). Migration required careful network planning, high‑availability design, security hardening, and fine‑grained cost accounting.

Key steps included:

Providing high availability across regions and availability zones.

Ensuring high performance by maintaining CPU utilization above 40‑50%.

Guaranteeing security with VPC isolation, ACLs, database audits, WAF, and DDoS protection.

Enhancing operational capability through resource accounting and precise billing.

Stateful services demanded complex migration: custom SDKs were built to bridge JD’s JimDB cache with public‑cloud Redis, and automated tools reduced DBA workload for migrating thousands of databases.

Applications were classified into three tiers (zero, first, second) based on business impact, with migration proceeding from low‑impact to high‑impact systems, employing gray‑scale testing and rollback plans.

03 First Major Promotion After Cloud Migration

In 2020, JD’s logistics completed full cloud migration, handling millions of orders daily on JD Cloud. The 618 promotion tested the hybrid‑cloud “cloud‑ship” platform, which orchestrated over 10 million CPU cores and 2 million Pods, achieving a three‑fold increase in resource utilization and a 30% reduction in per‑order cost.

During the 2021 Double‑11 event, JD faced additional challenges such as nationwide power‑saving measures and a shifted promotion start time (20:00). The cloud‑ship system dynamically allocated resources across public and private clouds, smoothing the pulse‑type traffic surge.

04 Conclusion

JD’s cloud journey illustrates that cloud adoption is no longer optional; it is essential for scaling complex logistics operations. Over five years, JD built a hybrid‑cloud ecosystem that delivers elasticity, cost savings, and resilience, positioning the company for continued digital transformation.