Observability in 2025: Scale, Traces, and AI

Observability's Next Frontier

This ClickHouse blog post delivers a comprehensive overview of the observability landscape in 2025, offering valuable insights into key trends and challenges. The article rightly identifies high cardinality, trace dominance, and data quality as critical issues. The emphasis on OpenTelemetry adoption is particularly noteworthy, as is the recognition of open-source observability's growing appeal due to cost predictability. The discussions around AI-driven SRE, while acknowledging the progress, realistically assess the current limitations, indicating a balanced perspective. The blog's focus on the practical implications, such as the shift towards data quality and usability, is a crucial takeaway for developers and organizations. However, the analysis could benefit from a more detailed exploration of the specific technical challenges associated with each trend. For example, a deeper dive into the architectural considerations for handling high-cardinality data or the specific performance bottlenecks encountered in trace-first workloads would enhance the value for a technical audience. Furthermore, while the article touches upon the limitations of AI-driven SRE, a more specific discussion about the types of AI techniques being employed and their respective strengths and weaknesses would be beneficial. The article also focuses primarily on the ClickHouse ecosystem, which is understandable, but expands to cover related projects such as HyperDX, Bindplane, and Odigos, but a wider range of comparison with other solutions would enhance its value.

Key Points

• High cardinality is the new bottleneck, demanding efficient storage and querying capabilities beyond simple ingestion.
• Tracing has become the primary signal, requiring platforms built with trace-first workloads in mind, and creating heavy sampling costs.
• OpenTelemetry adoption has accelerated, solidifying its place as the standard.
• Open source observability is gaining traction due to cost predictability and data ownership.
• AI-driven SRE is still in its early stages, with co-pilot-style workflows showing promise, but with significant limitations.
• Data quality and usability are emerging as critical considerations, shifting the focus from simply collecting telemetry to understanding which signals are useful.

---

📖 Source: Observability - a year in review