How a game studio cut its cloud bill 49% in 8 weeks

4 min read · Updated 18 August 2026

A mid-sized live-service game studio was spending about $2,600 a month on cloud infrastructure and suspected some of it was waste - but nobody had three weeks to find out. Over an 8-week engagement we cut the bill to $1,325/month: a 49% reduction, roughly $15,300 a year saved, with about 4 hours of their team's time and no impact on the live game.

The situation

Like most live-service teams, the studio had built infrastructure quickly to keep the game running and iterate on features. Cost was never the priority - and it shouldn't have been. But over a few years of shipping, the usual sediment had built up: services spun up and forgotten, resources that outlived the features that created them, and environments no one had turned off. The bill was survivable, but a meaningful slice of it was buying nothing.

What we found in five sessions

We took read-only access, pulled six months of usage, and ran our audit playbook across every service, region, and account. The costed findings list included:

• 24 dormant ECS services - quietly billing on Fargate → paused
• 6.4 TB of orphan EBS volumes - DeleteOnTermination=false leftovers → $318/mo
• 1 dormant MongoDB M10 cluster - paused-but-not-deleted → $107/mo
• NAT gateways across 4 regions - idle → deleted → $120/mo
• 18 orphan Elastic IPs - released across 7 regions
• Multiple orphan load balancers - deleted → $66/mo

How we did it

Nothing here was a rewrite. Each finding was a specific, reversible change, and the studio approved every one before we touched it. We worked in guided sessions: we'd walk through a batch of findings, explain the risk and the saving, and only act with an explicit go-ahead. Deletions were staged safely - snapshots first, deregister-before-delete, and a watch period before anything permanent - so nothing could take the live game down.

The result

Eight weeks in, the monthly bill was down 49% and holding, with no degradation to the game or its players. The studio came away with a documented record of exactly what changed and why, so the savings stick and a future FinOps hire could pick up where we left off. Total time asked of their engineers: about four hours.

This is a representative example of a first engagement. Every environment is different, but a never-optimized cloud bill almost always has a comparable slice of waste hiding in it - the question is just how much.