Observability vendor groundcover has put a feature called Autonomous Infrastructure into beta, according to SiliconANGLE. It automatically adjusts the processor and memory requests on customers' Kubernetes workloads, using how each workload actually behaves in production. The underlying technology comes from Wand Cloud, a resource-optimization startup that groundcover acquired in September.
The target is a familiar weak point. Engineers set requests for each workload up front, often before it has served real traffic, and groundcover says most teams rarely revisit them. A request set too low tends to surface as out-of-memory errors, while unused padding shows up on the monthly cloud bill.
Groundcover's sensor has already been collecting production history, so a customer on a supported version can open the feature without installing anything. The view places requested processor and memory next to actual usage, shows estimated monthly savings beside a count of under-provisioned workloads, and charts how much of the potential saving has been captured over time.
Dollar figures in the beta rely on a per-unit price for processor and memory that customers enter themselves. Public cloud pricing is planned next, followed by billing data such as the AWS Cost and Usage Report.
Groundcover argues that most optimization tools size workloads in isolation, which can let a request that suits one service starve its neighbors on a busy node. Before raising a request, its software checks how densely the node is already packed.
The feature is off until an engineer enables it and selects the deployments, namespaces or clusters it should manage. In the beta it changes requests only and leaves limits untouched. Each policy carries minimum and maximum bounds for processor and memory, and an audit log records who placed each workload under a policy and when. The company recommends starting with lower-risk targets such as development namespaces and avoiding something like a production database that needs headroom.
The beta is available at no extra cost to every customer running the groundcover sensor on Kubernetes, and the company is recruiting design partners. Demonstrations are planned for KubeCon + CloudNativeCon North America in Salt Lake City, Nov. 9-12. Groundcover has raised $160 million to date, including a $100 million Series C led by One Peak in July.
Why it matters: Overprovisioned Kubernetes requests are a direct source of wasted compute spend, and this release shows an observability vendor moving from reporting the gap to acting on it. Results are not yet reported, the beta changes only requests, and its savings figures depend on customer-entered pricing, so the real-world impact remains to be shown.
TwinStack automated analysis, based on reporting by SiliconANGLE. Check the original for full details.