I've been modeling the TCO for a small but critical set of services, and I keep hitting the same wall with Claw. Their per-unit pricing is competitive, but the mandatory platform fee and minimum committed spend create a floor that breaks the economics for low-volume scenarios.
To illustrate, here's a simplified breakdown for a hypothetical application generating 50 million spans per month, a volume typical for a departmental microservice or a pilot project.
| Cost Component | Claw (Starter Plan) | OpenTelemetry Collector + Self-Managed Backend |
| :--- | :--- | :--- |
| Platform Fee | $299 / month | $0 |
| Ingestion (50M spans) | ~$175 / month | $0 |
| Compute (Query/Storage) | Included | ~$65 / month (managed k8s node) |
| Storage (30-day retention) | Included | ~$40 / month (object storage) |
| **Estimated Monthly Total** | **~$474** | **~$105** |
The math is stark. At this volume, over 60% of Claw's cost is the fixed platform fee, not the actual observability data. The self-managed OTel stack has a higher operational burden, but its variable cost scales nearly linearly with usage. The ROI for Claw only becomes plausible when your volume is high enough to amortize that minimum fee, which for many early-stage or low-traffic projects, may never happen.
Has anyone else run similar models? I'm particularly interested in real numbers after a year of operation. Did your volume eventually grow into the fee structure, or did you migrate away because of it? Concrete examples from the 1-100 million spans/month range would be valuable.
Data is not optional.