Home/Library/Rightsize Compute Engine
How-to · Google Cloud · Updated May 2026

How to Rightsize Compute Engine VMs With Recommender

Most Compute Engine fleets are oversized, because instances are sized on launch-day guesses and never revisited. Google Cloud Recommender watches real utilization and tells you exactly which VMs to shrink. This guide shows how to read those recommendations and apply them without breaking anything.

Rightsizing Compute Engine VMs with Recommender means letting Google Cloud's machine-type recommendations do the analysis for you. Recommender, part of the Active Assist family, samples each instance's CPU and memory usage over time and proposes a smaller machine type when an instance is consistently underused. It is the fastest, lowest-risk compute saving on Google Cloud, and it needs no third-party tooling.

This how-to is part of our Google Cloud cluster. For the wider context, start with our complete guide to Google Cloud cost optimization, the pillar this piece links up to.

Step 1: find the recommendations

VM machine-type recommendations appear in the Compute Engine instances list, on each instance's details page, and in the Recommendation Hub. Each one shows the current machine type, the proposed type, and the estimated monthly saving. You can also pull them programmatically through the Recommender API or export them, which is how you analyze a large fleet at once rather than instance by instance.

Step 2: read the recommendation correctly

Recommender bases its proposal on observed peak and average usage over a recent window, usually the last several weeks. Check the observation period covers a representative load, including any monthly batch peaks, before you trust it. A recommendation generated during a quiet period can be too aggressive. Where a VM has predictable spikes the sampling window missed, widen the period or hold the instance back from the batch.

Step 3: apply changes safely

Changing a machine type requires stopping the instance, so treat it like a maintenance action: do it in a window, on non-production first, and one tier at a time rather than jumping several sizes. For managed instance groups, update the instance template and roll it out gradually. Keep the change reversible, and confirm the application is healthy at the new size before moving to the next batch. Custom machine types let you tune CPU and memory independently when no predefined type fits.

Step 4: clear idle VMs while you are there

Recommender also surfaces idle VM recommendations, instances that are running but doing almost nothing. These are not rightsizing candidates, they are deletion or suspension candidates. Confirm ownership, snapshot anything you might need, then stop or delete them. Idle cleanup often recovers more than rightsizing because an idle VM is 100 percent waste, not just oversized.

Want the whole fleet rightsized for you?

Our Google Cloud cost audit pulls every Recommender and idle finding across your projects, ranks them by dollars, and applies them safely. On the performance model, you pay only from realized savings. No savings, no fee.

Book a GCP cost audit →

Step 5: rightsize before you commit

The order matters. Rightsize the fleet first, then size your committed use discounts against the smaller, clean baseline. Buying a commitment on today's oversized footprint locks in the waste for one to three years, which is the most expensive mistake in Google Cloud cost. Once the VMs are right, read how sustained use discounts work for the automatic rate win, then committed use discounts explained for the deliberate one.

FindingActionRisk
Oversized VMMove to smaller machine typeLow, in a window
Idle VMStop, suspend, or deleteLow after ownership check
No predefined fitUse a custom machine typeLow
Predictable spikesWiden observation window firstAvoids under-sizing

Recommender and Active Assist behavior above reflects Google Cloud as of May 2026. Verify current product names and recommendation logic in Google Cloud documentation before acting, as the platform changes.

Go deeper · free guide

The Google Cloud Cost Optimization Field Guide includes the Recommender API queries and the rollout checklist behind this article. It is the downloadable companion.

The short version

Find the VM machine-type and idle recommendations in Recommender, check the observation window is representative, apply changes safely in a maintenance window starting with non-production, clear idle instances, and rightsize before you commit. It is the cleanest compute win on Google Cloud. For the full list of fast wins, see our Google Cloud checklist of 30 quick wins. When you want the fleet rightsized for you, that is what our Google Cloud cost optimization service delivers.

The Cloud Cost Brief

Cloud pricing moves. We tell you when it matters.

New commitment instruments, FOCUS changes, hyperscaler pricing shifts, and the plays that actually move a bill. No schedule, no filler.

Subscribe · Work email only