groundcover raised $100M for BYOC AI agent telemetry to stop data sampling

groundcover just raised $100 million to build “bring-your-own-cloud” observability for AI agents, and the bet is blunt: enterprises should never leave critical telemetry in someone else’s data plane. The round, led by One Peak, brings groundcover’s total funding to $160 million, and the company says it has more than 250 paying customers. In a market filled with “AI features” layered onto existing tools, groundcover is aiming at the plumbing itself, arguing that the assumptions behind traditional observability break once AI systems get autonomous.

At the center of that pitch is Shahar Azulay, groundcover’s co-founder and CEO, who says telemetry is exploding and users are frustrated by not getting “all the value” from Datadog and similar platforms. He points to teams limiting data, sampling traces, siloing it, and shortening retention to manage cost. groundcover’s response is a structural workaround: instead of storing customer telemetry inside groundcover-managed infrastructure, it keeps the data plane inside the customer’s own AWS, Microsoft Azure, or Google Cloud environment, while groundcover provides a managed control plane and user experience.

If this sounds like observability nerd stuff, the market signals say otherwise. The AI agent observability space is taking off, and groundcover is positioning itself as increasingly replacing established observability platforms inside enterprise environments. That matters because the incumbents are entrenched. Datadog, Dynatrace, New Relic, Splunk, and Grafana have years of product maturity and represent billions of dollars in annual revenue. Breaking into that group has “never been easy,” but groundcover is making a specific argument: artificial intelligence changes what observability has to do. It is not just an add-on anymore.

Traditionally, observability has been treated like post-production hygiene. Deploy software, monitor logs, metrics, and traces, investigate incidents, improve reliability over time. That workflow is shifting. AI-assisted software development accelerates deployment cycles. Coding assistants generate more code, infrastructure changes faster, and organizations run complex distributed systems that mix microservices, Kubernetes clusters, APIs, and large language models. Layer on AI agents, which can execute multi-step workflows, call external tools, and interact with production systems, and you get a telemetry explosion.

This is where the conflict gets practical. AI doesn’t just add more requests. It adds extra operational dimensions: prompt execution, model latency, token consumption, retrieval pipelines, tool invocations, and agent behavior. For enterprises experimenting with more autonomous systems, that data becomes valuable because it provides context for what the AI system actually did and why. But value collides with pricing, especially common models that charge based on telemetry ingestion volume. The result is often the same coping mechanisms: sampling traces, shortening retention, and limiting what gets collected. Those cost-control moves reduce spend, but they also reduce visibility exactly when autonomous systems need the most complete context.

groundcover’s “BYOC” approach is the core of its strategy and it is designed to change those incentives. In a conventional SaaS setup, the vendor typically stores customer telemetry in vendor-managed infrastructure. In groundcover’s model, customers keep the telemetry storage and processing inside their own cloud environment. A fully self-hosted deployment option is also available. The company notes that other vendors, including Datadog and a few others, may offer limited hybrid or customer-controlled data residency options, but it says those are not equivalent to a full BYOC model because telemetry is still processed and stored in the vendor’s managed infrastructure, with only partial controls like regional data residency, private links, or selective log forwarding.

That architectural difference influences pricing. groundcover argues it can avoid charging based on telemetry ingestion because customers are already paying for their own cloud infrastructure. Instead, it prices primarily on monitored hosts, regardless of telemetry volume. The company’s claim is that this changes customer behavior: teams can retain more complete telemetry for operational analysis, compliance, and AI-assisted troubleshooting rather than deciding what to cut because it is too expensive to keep. The important nuance is that host-based pricing will not be universally cheaper. groundcover’s own briefing says per-host pricing is most advantageous for organizations with high telemetry density, and may be less compelling for lightly utilized fleets. Still, the broader claim is about predictability. Enterprise infrastructure teams struggle with observability bills that fluctuate as applications grow, and groundcover is aligning costs with infrastructure planning rather than data generation.

The second pillar is technical differentiation built on eBPF. eBPF is a Linux kernel technology that has become a major building block for cloud observability. It can observe network traffic, system calls, and application behavior from inside the operating system kernel without requiring developers to manually instrument code. That helps reduce instrumentation complexity, which can shorten deployment times and improve telemetry coverage, especially for Kubernetes and cloud-native systems. Azulay argues this becomes even more critical as AI workloads create more complex interactions across services. His point is that groundcover’s “sensor” can observe systems deeply from infrastructure to application to AI workloads without developers needing to instrument code.

To be clear, eBPF is not unique to groundcover. Many observability vendors now incorporate it. What groundcover emphasizes is the combination: automatic eBPF collection paired with customer-controlled storage, OpenTelemetry compatibility, and what the briefing describes as unifying layers of observability across those data sources. The strategic implication for executives is straightforward. If AI agents keep generating more telemetry faster than infrastructure grows, then pricing and architecture will increasingly decide whether observability teams deliver full context or paper over gaps with sampling. The companies that win the next phase of observability may not be the ones with the best AI “assistant.” They may be the ones that get telemetry economics and data control right before enterprises reach the next retention cliff.