The $100 million bet isn't on better dashboards—it's on the assumption that when your agents outnumber your engineers, you can't afford to send their telemetry to someone else's cloud.
The Summary
- Observability startup groundcover raised $100M led by One Peak, bringing total funding to $160M with 250+ paying customers and 3x ARR growth
- Core thesis: AI agents produce so much telemetry that the centralized SaaS observability model (Datadog, Dynatrace, Splunk) breaks economically and architecturally
- The pitch is infrastructure, not features—keep agent telemetry in your own cloud, process it there, never send raw data out
The Signal
Observability has always been about watching what your software does after you ship it. Logs, metrics, traces. You deploy, something breaks, you dig through the data to figure out why. The business model that emerged around this—send all your telemetry to a vendor's cloud, pay by volume—worked fine when humans wrote the code and humans investigated the incidents.
That model assumes telemetry volume grows linearly with engineering headcount. AI agents break that assumption. An agent-assisted development pipeline might generate 10x the deployments. An autonomous agent handling customer service might spawn hundreds of parallel execution traces for a single user request. When your agents outnumber your engineers and operate 24/7, telemetry isn't a monitoring problem anymore. It's an infrastructure cost problem.
"AI-assisted software development has dramatically accelerated deployment cycles. Coding assistants generate more code, infrastructure evolves more rapidly."
groundcover's argument is simple:
- Traditional observability platforms charge based on data ingestion volume
- AI agents produce exponentially more telemetry than human-written systems
- Sending all that data to a third-party SaaS cloud gets prohibitively expensive
- The solution is to process telemetry in your own cloud and only send summaries out
This isn't just about cost. It's about control. When you're running autonomous agents that touch customer data, execute financial transactions, or make operational decisions, the telemetry itself becomes sensitive. Do you really want every trace of every agent decision leaving your infrastructure? groundcover is betting enterprises will increasingly answer no.
The company claims it's replacing Datadog, Dynatrace, New Relic, Splunk, and Grafana inside enterprise environments. Those are platforms with billions in combined revenue and years of product depth. groundcover is four years old. The fact that it's gaining traction suggests something structural is shifting, not just that they built a better dashboard.
The Implication
If groundcover is right, the observability market is about to fragment. Traditional platforms won't disappear, but they'll increasingly serve human-scale systems while a new generation of tools handles agent-scale telemetry. The strategic question for enterprises isn't just "which observability platform?" but "where does our telemetry live and who controls it?"
For anyone building or deploying AI agents at scale, this is worth tracking. The telemetry problem is real. If your agent fleet grows faster than your budget for third-party SaaS, you'll need a different architecture. Whether groundcover is the answer or just the first credible attempt at one, the problem they're solving isn't going away.