The economic divergence of observability in 2026

Datadog pricing multiplies across independent meters for infrastructure, APM, logs, and custom metrics. The cost of running a mid-size fleet of fifty hosts with infrastructure monitoring and APM, five hundred gigabytes of monthly log ingest, and two hundred custom metrics exceeds two thousand eight hundred dollars per month on Datadog. This cost compounds because the per-host structure doubles infrastructure and APM expenses every time you add a host. Grafana Cloud avoids this specific host multiplier by basing costs on metrics series count and telemetry volume. 95th percentile billing protects users from unexpected spikes. You can choose to self-host the LGTM stack to eliminate software license fees entirely. This choice replaces the vendor invoice with the cost of engineering time spent running four separate systems.

Log indexing and search architecture

Kibana dominates log analysis because its full-text search handles billions of lines using Lucene. This approach demands significant RAM and compute to manage inverted indices. Grafana Loki uses a different strategy. It indexes only structural metadata like labels and tags. It keeps the raw log content in cheap object storage. This makes Loki significantly more cost-efficient for high-volume retention than Elasticsearch. Datadog also uses full-text indexing but adds separate costs for log ingestion and indexing.

Datadog’s pricing structure introduces significant forecasting uncertainty for growing teams. A single feature like LLM observability triggers a $120 per day charge without user confirmation when it detects LLM spans. An AI feature deployment increases your monthly bill by $3,600 automatically. You must also manage custom metric costs, where each unique tag-value combination can create a new billable metric. Datadog includes 100 custom metrics per host in the Pro tier, but additional metrics cost money.

The decision between a modular stack and a unified platform depends on whether a team prioritizes flexibility or ease of use. Teams migrate from Datadog to Grafana when the SaaS bill exceeds the cost of hiring one or two dedicated observability engineers. These engineers cost roughly $150,000 each per year. Conversely, teams move to Datadog when they lack the bandwidth to manage the four separate components of the Grafana stack. If you require full-text search, do you prefer to pay for high RAM requirements or for managed indexing services?

The unified platform and the cost of cardinality

OpenObserve uses a Parquet-based columnar format on object storage to cut costs. In a test run using the OpenTelemetry Demo, Datadog cost $174 per day while OpenObserve cost only $3 per day. This 58x difference exists because OpenObserve lacks the per-host tax and the premium charges for specific telemetry types. Datadog charges $36 per month per APM host in the Pro tier. It also charges $18 per month per infrastructure host. These charges scale with your container count, which penalizes horizontal scaling during traffic spikes.

OpenObserve provides deployment choices that neither incumbent offers together. You can self-host for full data control or run OpenObserve Cloud managed with the same usage-based pricing model. OpenObserve provides unlimited custom metrics with no per-metric charges. This removes the need to hunt down high-cardinality metrics to avoid financial penalties. Datadog creates friction because users limit instrumentation to control costs. Users often restrict tagging or use conservative sampling to avoid massive bills. OpenObserve treats all telemetry as standard data. It does not charge extra for per-span fees or for specific AI-related spans. OpenObserve also uses a single query language, standard SQL, across logs, metrics, and traces.

Netdata provides another alternative with a per-node pricing model. This model avoids the cardinality trap that hits both Grafana and Datadog. Netdata Cloud Business starts at $4.50 per node per month on annual plans. This price does not change based on metrics volume, users, or data retention. You can deploy Netdata as a single-binary agent with one command. If you already manage a large Kubernetes fleet, you might find that the per-host multipliers of traditional SaaS platforms make granular observability impossible to budget for.