How is your enterprise tracking AI agent telemetry? Groundcover thinks it should never leave your cloud

Groundcover just raised $100 million at a valuation that implies the observability market is ready to fracture. One Peak led the round. Total funding now sits at $160 million. The company claims 250 paying customers and tripled annual recurring revenue in twelve months. Those are self-reported figures. But they land in a market that has resisted disruption for a decade. Datadog, Dynatrace, New Relic, Splunk and Grafana collectively generate billions in revenue. Their products are mature. Their contracts are sticky. Breaking in has never been easy.

Groundcover is not trying to win a feature war. It is arguing that the architecture underneath observability is obsolete. The thesis: AI has rewritten the assumptions those platforms were built on. If that thesis holds, the incumbents are not vulnerable to better dashboards. They are vulnerable to a different data gravity model.

Observability has always been a post-production discipline. Engineers deploy, then monitor. Logs, metrics, traces flow to a centralized platform. Incidents get investigated. Reliability improves over cycles. That workflow assumed human-authored code, predictable deployment cadences, and telemetry volumes that grew linearly with infrastructure. AI-assisted development shattered all three.

Coding assistants now generate more code in a week than teams used to ship in a quarter. Infrastructure evolves daily. Distributed systems combine microservices, Kubernetes clusters, APIs and large language models into topologies that change before documentation catches up. Meanwhile, enterprises are deploying AI agents that execute multi-step workflows, call external tools, and mutate production state autonomously. Every step emits telemetry.

The volume explosion is not the only problem. The character of the data has changed. Traditional infrastructure monitoring captures CPU, memory, latency, error rates. AI applications demand observability into prompt execution, model latency, token consumption, retrieval pipelines, tool invocations, and agent decision traces. That telemetry is not noise. It is the only audit trail for autonomous behavior. When an agent makes a wrong call, the context lives in the prompt chain, the retrieval scores, the tool response — not in a stack trace.

Enterprises increasingly want to retain all of it. Discarding data that explains why an autonomous system acted is a liability. But the dominant pricing model charges by ingestion volume. Retention becomes a budget crisis. Engineers respond by sampling, by dropping traces, by narrowing windows. They blind themselves to the very systems they are trying to control.

Groundcover's architecture answers that tension by refusing to move telemetry off-premises. The platform runs inside the customer's cloud account. Data never leaves the VPC. Compute scales with the customer's own elastic infrastructure. Ingestion costs collapse to the marginal price of the customer's own storage and compute. There is no vendor meter spinning on every span.

This is not a new idea. Security and compliance teams have demanded data residency for years. What is new is the scale argument. AI telemetry volumes make SaaS egress fees punitive. They make network transfer a latency tax on real-time agent observability. They turn the observability pipeline into a bottleneck that slows the very systems it is meant to illuminate.

The incumbents know this. Datadog and Dynatrace both offer on-premises or private-link deployments. But those are adaptations of architectures designed for centralization. Their query engines, their retention tiers, their billing logic — all assume a SaaS control plane. Groundcover was built from the opposite assumption: the control plane follows the data. The query engine runs where the data lives. The customer owns the compute bill.

That architectural purity is the investment thesis. One Peak is betting that the next ten years of enterprise software will be defined by autonomous agents operating inside private clouds, and that the observability layer for those agents must be local by default. If they are right, Groundcover becomes the default substrate for AI operations. If they are wrong, the company is a niche vendor with an expensive architecture and a shrinking addressable market.

The skepticism writes itself. Two hundred fifty customers is a rounding error for Datadog. Tripled ARR from a small base is momentum, not scale. Enterprise procurement cycles are long. Security reviews are longer. Ripping out an observability platform that touches every service, every team, every compliance audit is a career-risk decision for a VP of Engineering. Groundcover will need to prove migration is survivable, not just desirable.

There is also the platform breadth question. The incumbents offer logs, metrics, traces, profiling, synthetics, RUM, security, CI visibility, cost allocation, and a marketplace of integrations built over fifteen years. Groundcover covers the AI-native surface exceptionally well. It is less clear how it handles the legacy surface that still generates 80 percent of enterprise telemetry. A dual-vendor strategy — Groundcover for agents, Datadog for everything else — is the path of least resistance. That outcome caps Groundcover's total addressable market at the AI slice.

But the slice is expanding. Every enterprise board is demanding an agent strategy. Every CTO is authorizing budget for autonomous workflows. The telemetry those agents produce will not fit the old pipes. The pricing models will not absorb the volume. The compliance requirements will not allow the egress. Groundcover does not need to replace the incumbents everywhere. It only needs to become the mandatory layer for the new workload. The incumbents can keep the rest.

The $100 million buys time to prove that mandate. It funds the sales motion to reach the next thousand enterprises. It funds the engineering to close the legacy gaps. It funds the narrative that observability is no longer a monitoring problem — it is an infrastructure problem, and infrastructure belongs to the operator.

The market will decide whether that narrative survives contact with procurement. But the fact that serious capital just bet nine figures on it means the conversation has shifted. The question is no longer whether AI changes observability. The question is which architecture wins when it does.