For your constraints (500 engineers, global SaaS, German data-localisation, telemetry must stay in your VPC, OpenTelemetry-first, >13-month retention), I would narrow the field to self-managed enterprise observability platforms, not the usual SaaS APM vendors.

The key architectural question is whether you want:

• A single integrated APM/observability product (simpler operations, usually higher licensing cost), or

• A composable OpenTelemetry-native stack (more engineering ownership, much better cost control at high volume).

Shortlist: self-hosted platforms that fit

Vendor / platformFully self-hosted in your VPCOpenTelemetry supportLong retentionCost predictabilityMy rankingGrafana Labs LGTM stack (Grafana + Loki + Tempo + Mimir)YesExcellentYes (you control storage lifecycle)⭐⭐⭐⭐⭐#1Elastic Elastic Observability self-managedYesExcellentYes⭐⭐⭐#2OpenSearch Project Observability StackYesNativeYes⭐⭐⭐⭐⭐#3Splunk self-managedHistorically yes (Splunk Enterprise)GoodYes⭐⭐#4DatadogNo (primarily SaaS)ExcellentDepends on contract⭐Not suitable

1. Grafana Labs LGTM stack — highest recommendation

For your stated priorities, this is the strongest fit.

The stack:

• Metrics: Grafana Mimir

• Logs: Grafana Loki

• Traces: Grafana Tempo

• Dashboards/exploration: Grafana

It is designed around OpenTelemetry ingestion and lets you keep all data inside your infrastructure. Mimir is built for horizontally scalable long-term Prometheus-compatible metrics storage, including very large deployments. Grafana Labs Loki supports configurable retention policies and object-storage-backed deployments, which is important when retention goes beyond a year. Grafana Labs (+1 more citation)

Why I rank it #1 for cost predictability

Your biggest cost problem is probably not licenses — it is telemetry cardinality and retention economics.

With LGTM:

• You pay for:

compute

• storage

• operations

• You are not paying per:

host

• engineer

• GB ingested

• indexed event

• trace span

That matters enormously at your scale.

A realistic architecture:

Applications | OpenTelemetry Collector | +--> Mimir (metrics) | +--> Loki (logs) | +--> Tempo (traces) | Grafana

For a 500-engineer SaaS company, I would expect this to beat commercial APM pricing once ingestion gets large.

Tradeoff: your SRE team owns the platform.

2. Elastic Observability — best "traditional APM" alternative

Elastic is probably the closest replacement for a commercial APM experience while staying self-hosted.

Strengths:

• Mature APM UX

• Strong service maps

• Good logs + traces correlation

• Strong search capabilities

• OpenTelemetry support through Elastic's OpenTelemetry distribution and ingestion paths elastic.co

Elastic supports self-managed APM deployments where components can run inside your own environment. Elastic

Where it loses against Grafana:

• Elasticsearch storage economics can become painful with very high-volume logs.

• Cost tends to follow indexed data volume.

• Long retention requires disciplined

Given your constraints, the field narrows considerably:

• Telemetry must remain entirely within your VPC/on-prem (German data localization)
• Full-stack observability (metrics, logs, traces, profiling is a bonus)
• Native OpenTelemetry support
• >13 month retention
• ~500 engineers with globally distributed teams
• Large-scale ingest, where cost predictability matters as much as features

The mainstream SaaS leaders (Datadog, New Relic, Dynatrace, Honeycomb, etc.) are effectively eliminated because they require telemetry to leave your environment (even if some offer regional hosting).

The strongest fully self-hosted options

My ranking for your environment

1. Grafana Enterprise Stack (LGTM)

This would probably be my first recommendation.

Typical deployment:

• Prometheus-compatible metrics (Mimir)
• Loki for logs
• Tempo for traces
• Grafana dashboards
• Pyroscope for continuous profiling

Advantages:

• Designed around OpenTelemetry
• Proven at enormous scale
• Storage can live in S3-compatible object storage inside your own cloud
• Multi-region deployment is mature
• Retention is simply storage policy—not a licensing restriction
• No per-host billing
• No ingest surprises if you're operating it yourself

Tradeoffs:

• Operational complexity
• You're running four major distributed systems
• Needs an internal observability/platform engineering team

Many organizations larger than yours successfully operate this architecture. middleware.io

2. Elastic Observability

Especially attractive if you're already invested in Elastic.

Pros:

• Mature search
• Excellent log analytics
• Strong APM
• Good OpenTelemetry support
• Very good RBAC
• Excellent long-term retention using searchable snapshots/object storage

Downside:

• Infrastructure can become expensive at very large ingest rates.
• License costs become meaningful beyond the free tier.

3. SigNoz

This has matured significantly.

Pros:

• OpenTelemetry-native
• ClickHouse backend
• Single integrated UI
• Easier than operating the full Grafana stack
• Good developer experience

Compared to Grafana:

• Simpler
• Less flexible
• Smaller ecosystem

For a 500-engineer organization, it's a realistic option. Community discussions frequently compare it with LGTM as one of the leading self-hosted OTel backends. www.reddit.com

4. OpenObserve

Very compelling if your biggest problem is the observability bill.

Strengths:

• Rust implementation
• Columnar storage
• Very efficient resource usage
• Unified logs/metrics/traces
• Good OpenTelemetry support

It's younger than Grafana or Elastic but is attracting organizations focused on reducing observability infrastructure costs. cubeapm.com

5. Uptrace

Excellent engineering design.

Especially appealing because:

• OpenTelemetry-first
• Predictable pricing model
• Easy self-hosting
• Unified experience

I'd choose it for simplicity, although I have less confidence than Grafana or Elastic for very large enterprise deployments. uptrace.dev

Cost predictability ranking

For organizations where observability costs have already become painful:

For your specific requirements

Given:

• 500 engineers
• Germany data localization
• OpenTelemetry
• Long retention
• Existing observability spend is a major concern

I'd evaluate these in order:

1. Grafana Enterprise Stack
2. SigNoz
3. OpenObserve
4. Elastic Observability

I would probably avoid platforms whose economics are primarily based on telemetry ingestion volume if your current pain point is an observability bill growing with data volume.

One architectural approach I'd also recommend, regardless of backend, is to place OpenTelemetry Collectors between your services and the observability platform. That gives you centralized control over sampling, filtering, enrichment, and routing. In practice, many organizations reduce telemetry volume substantially by filtering high-cardinality attributes, applying tail-based trace sampling, and routing only high-value logs to long-term storage, while preserving the option to change backends later because applications continue emitting standard OTLP telemetry.