Cloud Daily edition
CLOUD KUBERNETES AND PLATFORMS
China normalises cloud-native without abandoning the data centre
CNCF and SlashData estimate 1.75 million cloud-native developers in China, with growing adoption among young backend profiles. The most interesting finding is not a uniform shift to the public cloud, but the coexistence of Kubernetes, internal platforms, industry needs, and a strong private infrastructure base.
A new report from CNCF and SlashData puts the number of cloud-native developers in China at 1.75 million in the first quarter of 2026; some 400,000 also work in artificial intelligence. Among those who develop backend services in the country, the proportion classified as cloud-native has reportedly risen from 30% to 48% in two years. For developers under 25, this figure reaches 58%. The simple interpretation would be to proclaim that Kubernetes is now the standard. The more useful analysis is more precise: certain distributed practices are being incorporated into routine work, but on infrastructure and in sectors very different from those in the West.
The report defines a developer as cloud-native if they report using several technologies from a list that includes containers, orchestration, Kubernetes, functions, event-driven architecture, observability, immutable infrastructure, service mesh, chaos engineering, messaging, and multi-cluster management. For backend developers, it requires at least three; for other profiles, two. The data comes from an online survey of more than 12,500 people from 95 countries, conducted between December 2025 and January 2026 and weighted to correct for bias. The Chinese backend sample consists of 319 responses. This is sufficient to observe trends, not to treat each percentage as an exact market inventory.
On-premises servers and private cloud ahead of hybrid
This adoption does not mean abandoning the data centre. In China, on-premises servers remain the dominant deployment option and private cloud outweighs public cloud; hybrid cloud use is considerably lower than the global average. The report links this to an ecosystem of domestic providers that offers scale without some of the sovereignty trade-offs that push other regions towards a mix of public and on-premises cloud. Furthermore, manufacturing, telecommunications, hardware, and energy account for a larger fraction of technical employment. In this environment, cloud-native describes how software is packaged, deployed and observed, not necessarily where it runs.
The study provides another signal: 88% of global backend developers now work with some form of infrastructure standardisation, while the group with no formalised practice is shrinking. This may lead a programmer to report using fewer containers or less Kubernetes because they consume an internal platform without seeing its underlying components. In the association analysis, feature flags and observability act as bridges to more advanced practices. Immutable infrastructure, chaos engineering, service mesh, and multi-cluster management appear later in a closely related group. For AI teams, immutability in particular acts as a gateway to reproducible training and distributed inference.
The factory as a testbed: golden path before more layers
Industry illustrates this difference. 48% of professional Chinese IIoT developers fit the cloud-native definition, compared to 42% globally. An assembly line or a power grid requires continuity, local control, and long infrastructure cycles, but also reproducible builds, coordinated deployments, and common telemetry. Kubernetes and declarative patterns can operate in a factory, on a private edge, or in a national cloud. The operational advantage appears when they reduce variation between plants and allow a service to be restored; it disappears when they add layers that the team does not know how to diagnose or when a central update ignores production windows.
Practical application is not about buying the entire catalogue. An industrial team can start with a golden deployment path: signed images, declarative configuration, common observability, resource limits, a tested rollback procedure, and a feature flag to separate deployment from activation. For AI workloads, this adds model and data versioning, GPU quotas, and cost per inference. Only when SLOs, an owner, and a recovery procedure are in place does it make sense to evaluate service mesh, chaos engineering, or multi-cluster management. Each layer must respond to an observed failure and have an assigned operational cost; otherwise, maturity becomes showcase architecture.
Important questions remain. The survey is self-reported, the definition of backend has changed, and some apparent drops in usage are explained by methodology, not by technological abandonment. The associations show tools that appear together, but do not prove that one causes the adoption of another. It will be necessary to monitor the next wave of data, the evolution of hybrid and public cloud in China, and whether this growth translates into fewer incidents, better security, and more disciplined FinOps. The report confirms that cloud-native is spreading; it does not prove that all the organisations that name it know how to operate it.
Tags
- CNCF
- Kubernetes
- Cloud-native
- China
- IIoT
- Private cloud
BOLDERROR Daily edition Rubén Campoy