Observability startup groundcover raised $100 million in a Series C led by One Peak, bringing total funding to $160 million. The company reports 250-plus paying customers and tripled ARR over the past year, and it is targeting direct displacement of Datadog, Dynatrace, New Relic, Splunk, and Grafana inside enterprise environments. Its core argument is not that those platforms are bad, but that their foundational architecture predates the AI agent era and cannot survive it.
The technical case rests on two pillars. First, a bring-your-own-cloud model keeps the entire telemetry data plane inside the customer's AWS, Azure, or GCP environment, allowing groundcover to price by monitored host count rather than data ingested. This matters because AI agents, prompt pipelines, token consumption logs, and tool invocations inflate telemetry volume far faster than host count grows, making ingestion-based pricing increasingly punishing. Second, groundcover uses eBPF, the Linux kernel technology, to instrument applications without requiring code changes, which compresses deployment time and widens telemetry coverage across Kubernetes clusters running mixed AI and traditional workloads.
The piece is worth reading in full for one reason: it does not let groundcover off easy. The original reporting surfaces the conditions under which host-based pricing actually loses, specifically lightly utilized fleets with low telemetry density, and it draws a clear line between genuine BYOC architecture and the partial data-residency controls that competitors market as equivalent. If your organization is currently sampling traces or shortening retention windows to manage observability costs, the architectural tradeoffs described here are directly relevant to decisions you are likely making right now.
[READ ORIGINAL →]