The problem
Bullieverse was launching Necrodemic, a Web3 multiplayer title with players in NA, EU and APAC who all needed sub-100ms gameplay. Statically provisioned GPU fleets sat under 40% utilised off-peak and still buckled during 5–10× event spikes, while DDoS exposure and a small ops surface made every launch event a risk.
What we shipped
We rebuilt the platform on AWS GameLift with FlexMatch rules that balance skill, latency and capacity, and codified every region in Terraform modules so new geographies stand up in hours. Game telemetry streams into Amazon Redshift for live retention and cost dashboards, and a multi-layer auto-scaling policy mixes spot and on-demand capacity using player-queue depth and GPU utilisation rather than naïve CPU thresholds.
The outcome
GPU spend dropped 40% through right-sizing, GameLift spot integration and predictive scaling. Player latency landed under 80ms for 95% of sessions across three regions and matchmaking times fell 60%. The launch event drove a 7× traffic spike that the platform absorbed in under four minutes with zero manual intervention.