The move from AI experimentation to production is changing how enterprises design and operate their cloud infrastructure. A new Q3 2026 technology report from the Cloud Native Computing Foundation (CNCF), developed with developer research firm SlashData, finds that platform engineering ownership is closely associated with how organizations build and manage AI infrastructure. The research also identifies Prometheus, OpenTelemetry, OpenTofu, OpenFeature and Carina among the technologies developers consider mature, useful or reliable enough for broader adoption.
Platform Engineering Emerges as a Strategic Factor in Enterprise AI
The findings arrive as enterprises move beyond isolated AI pilots and begin integrating machine learning, generative AI and agent-based workflows into production environments. That transition introduces infrastructure requirements that differ from conventional application workloads, including specialised compute, data pipelines, model-serving systems, storage capacity and continuous monitoring.
CNCF’s research suggests that the way organizations structure platform engineering teams influences how they respond to these demands. Developers at organizations where individual teams independently manage their tooling are more likely to use hybrid AI infrastructure approaches than those working in organizations with custom internal developer platforms.
The difference is substantial: 54% of developers in independently managed environments reported hybrid AI approaches, compared with just 5% at organizations building custom internal platforms. Organizations where multiple teams collaborate on platform capabilities recorded a 43% hybrid adoption rate.
The report also found that 38% of developers at organizations building custom internal developer platforms use a separate, dedicated AI platform, while 33% extend their existing developer platform to support AI workflows. At organizations where platform teams curate third-party tools, 25% rely primarily on AI vendor platforms.
These results suggest that platform ownership is not simply an organizational design decision. It can influence whether companies consolidate AI infrastructure, extend existing engineering systems or combine internal platforms with external services.
Centralized platforms versus hybrid AI environments
A centralized platform engineering team can establish common standards for infrastructure provisioning, security, deployment and observability. This can make it easier to coordinate AI workloads across departments and reduce the operational overhead of supporting different tools.
However, centralization does not automatically mean every AI workload should run on one platform. Dedicated AI environments may be necessary for specialized compute, model development or inference workloads. Other teams may need to connect external AI services with existing enterprise systems.
The CNCF findings point to different strategies rather than a single universally preferred architecture. Organizations building custom internal developer platforms appear more likely to make a deliberate choice between extending existing infrastructure and creating a separate AI platform. Teams with more distributed ownership are more likely to combine approaches.
For enterprise technology leaders, the practical question is how to maintain consistent governance and operational visibility without limiting the flexibility needed by AI development teams.
Prometheus and OpenTelemetry lead observability
The report’s observability findings highlight the continuing importance of monitoring and telemetry as cloud native infrastructure becomes more complex.
Prometheus and OpenTelemetry both earned the report’s “Adopt” position. Prometheus received four- or five-star maturity ratings from 80% of surveyed developers, including 38% who awarded it the highest rating.
OpenTelemetry received five-star usefulness ratings from 34% of developers, while 92% of current and former users said they were likely or highly likely to recommend it. The results reinforce the importance of standardised telemetry for understanding application behaviour across distributed environments.
Other observability technologies also attracted strong recommendations. Fluent Bit recorded a 99% recommendation rate among current and former users, while Jaeger received a 93% recommendation rate. Cortex led the category for five-star maturity ratings, at 40%.
These measures capture different aspects of developer experience. A tool can receive strong recommendations without leading every maturity or usefulness measure. Fluent Bit’s results, for example, suggest strong user satisfaction even though its ratings for maturity and usefulness were more moderate.
For production AI systems, observability extends beyond conventional infrastructure monitoring. Teams need visibility into model-serving performance, application dependencies, resource consumption, latency and the behaviour of distributed services. Tools such as Prometheus and OpenTelemetry can contribute to that foundation, although they do not by themselves provide every capability required for AI-specific monitoring.
Infrastructure as code and storage remain foundational
The radar also evaluates infrastructure as code and storage technologies that support cloud native environments.
OpenTofu earned an “Adopt” position in the infrastructure-as-code category, with 96% of current and former users saying they were likely or highly likely to recommend it. The tool provides an open-source infrastructure-as-code workflow for defining and managing cloud and infrastructure resources.
OpenFeature also received an “Adopt” position in the report’s infrastructure-as-code grouping, with 67% of developers giving it four- or five-star maturity ratings and 29% awarding it five stars for usefulness.
There is an important technical distinction here: OpenFeature is a vendor-neutral standard and API for feature flagging, not a conventional infrastructure-provisioning tool. Its inclusion in the report’s IaC category should not obscure its actual function. It provides a consistent interface for integrating feature-flagging systems across different providers. OpenFeature+2
In storage, Carina earned the “Adopt” position. Rook and Longhorn also received strong developer recommendations, with 94% of Rook users likely or highly likely to recommend it and 59% of Longhorn users highly likely to do so.
Storage technologies are particularly relevant to AI infrastructure because data-intensive workloads depend on reliable access to datasets, model artefacts and application state. Their operational characteristics can affect how organizations manage distributed workloads and scale services across clusters.
Developer confidence is not the same as project maturity
One of the report’s important distinctions is that its “Adopt,” “Trial” and “Assess” categories reflect developer perceptions. They are not equivalent to CNCF’s formal project maturity levels of Sandbox, Incubating and Graduated.
The distinction matters. OpenTofu, for example, received strong developer recommendations despite being a CNCF Sandbox project. OpenFeature is an Incubating project. The report’s radar positions instead combine developer assessments of maturity, usefulness and willingness to recommend a technology.
The research surveyed more than 300 professional developers working with cloud native technologies. The report describes technology-specific respondent groups, so individual percentages should be interpreted within the relevant category rather than as results from one identical sample.
For enterprise leaders, the broader message is that AI infrastructure is becoming a platform engineering challenge as much as a model-development challenge. Organizations need systems that can support experimentation while maintaining the reliability, visibility and consistency required for production operations.
CNCF’s Q3 findings do not establish that one platform architecture will work for every enterprise. They do, however, show how engineering ownership and confidence in foundational cloud native tools are influencing the infrastructure decisions companies make as AI adoption expands.
Market Landscape
The cloud native market is moving from container orchestration and application modernisation toward a broader infrastructure layer for AI development and production deployment.
Three trends stand out:
- Platform engineering is becoming central to AI operations: Enterprises are deciding whether AI workloads should extend existing internal developer platforms, run on dedicated AI infrastructure or use hybrid architectures.
- Observability is a production requirement: Monitoring, metrics, logging and distributed tracing help engineering teams understand complex applications and infrastructure as AI workloads scale.
- Infrastructure standardisation remains important: Infrastructure-as-code tools, feature-flagging standards and cloud native storage systems help teams manage configuration, deployment and operational consistency.
The CNCF report is useful because it connects technology preferences with organizational structure. It suggests that infrastructure choices are influenced not only by technical requirements but also by who owns platform capabilities and how teams collaborate.
The wider ecosystem includes Kubernetes, Backstage and cloud infrastructure services from Amazon Web Services, Microsoft Azure and Google Cloud. These technologies and services are part of the broader platform landscape in which enterprises are building and operating AI systems.
The competitive question is shifting from whether a company can deploy AI to whether its infrastructure can support AI reliably, securely and efficiently across teams and workloads.
Top Insights
- CNCF and SlashData found that platform engineering ownership is closely associated with how organizations choose to build, extend or combine AI infrastructure.
- Hybrid AI workflows were reported by 54% of developers in independently managed tooling environments, compared with 5% at organizations building custom internal platforms.
- Prometheus and OpenTelemetry earned “Adopt” positions in observability, reflecting strong developer perceptions of maturity, usefulness and reliability.
- OpenTofu recorded a 96% recommendation rate, while Carina earned the report’s “Adopt” position in cloud native storage.
- The report’s adoption categories reflect developer perceptions and should not be confused with CNCF’s formal Sandbox, Incubating and Graduated project maturity levels.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
