GameFabric or GameLift? What Studios Gain From One Fleet Across Bare Metal and Cloud
Whether you're picking an orchestration platform before launch or considering alternatives to GameLift, your server infrastructure decision comes down to three questions: what capacity you can run, how many game servers each instance carries, and how fast the fleet reacts when players arrive.
GameLift's managed fleets run in the AWS cloud. Bringing your own hardware is possible, but AWS lists that option as self-managed. So your baseline either sits on cloud capacity, or your team takes on the hardware.
GameFabric removes that trade-off: your predictable baseline on cost-efficient dedicated hardware, your peaks in the cloud you've already committed to, and none of the stitching left to your engineers. We run both as one fleet, size every game server by what it actually needs, and ship the observability, DDoS protection and open tooling your team would otherwise build and manage itself. Behind it, our SRE team sources and operates the bare metal, scales the clusters and keeps the platform running, so your engineering time and budget go into your actual game. Here's how that works in practice.

Already committed to AWS? Want to go multi-cloud but don’t want the headache? Bring Your Own Cloud (BYOC), where GameFabric manages game server capacity inside your own cloud account, runs on the instances you already pay for. AWS keeps billing you directly, so eligible committed spend and negotiated rates can continue to apply, subject to your provider agreement, or hand it over to us and we run the account and the bare metal.
Layer | Owner |
|---|---|
AWS account, IAM, quotas, billing, committed rates | You |
Locations, clusters, node scaling | GameFabric SRE team |
Fleet configuration | You, in Terraform, OpenTofu, the REST API or the UI |

Fixed instance sizing reserves capacity for the worst moment. GameFabric sizes for the guaranteed minimum and lets each game server borrow the rest.
Feature | What it does |
|---|---|
Bare metal baseline | Your predictable CCU fills dedicated hardware first, at priority 0, and cloud only takes the overflow |
Requests and limits | Each game server gets a guaranteed CPU and memory baseline, defined by you. When more resources are required for short periods, your game servers can borrow from the host, allowing more servers to be packed onto a single instance. |
Packed scheduling | The default strategy co-locates game servers to maximize density. |
Shared Maximum Replicas | Intentional overcommitment across Armadas, our session-based server fleets, sharing a Location, since not every server is Allocated at once |
Dynamic Buffer | Sizes the warm pool from live demand, with a cost-versus-availability slider |
Scale to Zero | Auto scales lower-priority cloud capacity down when there’s no demand, for session fleets with at least two Region Types. Fixed cloud infrastructure keeps running |
Fenris Creations (formerly CCP Games) built EVE Vanguard's fleet on this model. Each game server starts with the minimum it needs, draws on the host's spare capacity when matches get busy, and gives it back afterwards. The result: more sessions on each instance than they historically ran.

Within each Region, you group capacity into Region Types, such as bare metal and cloud, and give each one a priority. GameFabric fills the priority 0 type first and only spills over to the next when it's full.
Capacity | Role | Max game servers per node | Billed by |
|---|---|---|---|
Bare metal, operated by GameFabric | Daily baseline, priority 0 | 150 | GameFabric |
BYOC | Peaks, against your commitment | 100 | Your Cloud Provider(s) of Choice |
GameFabric Cloud | Elastic capacity with no cloud account to manage | 100 | GameFabric |

Capacity is half the platform. The rest is what your team stops needing to build and maintain.
Capability | On GameFabric |
|---|---|
Provisioning | 250+ game servers per second |
Observability | Grafana, Prometheus and Loki out of the box, plus eBPF CPU profiling at 2 to 3% overhead, with no game server code instrumentation |
Session and long-running servers | Armadas for matches, and Vessels, individual named game servers, for long-running worlds. Vessels survive updates and move between bare metal and cloud without being recreated |
DDoS protection | SteelShield™ Level 3 Proof of Identity ties every packet to an authenticated account, with 0% false positives for titles that complete the integration |
Open standards | Agones and Kubernetes underneath, fleets as code in Terraform or OpenTofu, and any matchmaker that can make one REST call |


Fenris Creations ran EVE Vanguard's session infrastructure with four backend engineers. By building on GameFabric, they saw 60% lower server hosting costs compared with their previous provider. When the Operation Nemesis playtest needed more servers, GameFabric added the capacity in under 30 minutes.

If you host on Amazon GameLift today, your matchmaker can stay. FlexMatch switches from managed to standalone mode, and an AWS Lambda function replaces the managed queues: it requests a ready server from the Allocator, our managed broker between matchmakers and ready servers, and attaches the connection details to the match event. Your game client and backend see configuration and variable changes. Your game server swaps lifecycle calls.
Phase | GameLift server SDK | Agones on GameFabric |
|---|---|---|
Startup |
|
|
Health | Responds to | Sends |
Allocation |
| Watches for the allocation trigger, no confirmation call |
Shutdown |
|
|
Two things to check first: the path may not work if your game service relies on GameLift game sessions, and the Allocator is ordered separately.

Same game, fewer engineers tied to server infrastructure, and a fleet sized to what your players actually do. Whether you're choosing a platform or reconsidering the one you're on, bring your hourly player curves per region, players, CPU and memory per game server, and any cloud commitments to a demo. Already on GameLift? Start with the FlexMatch integration guide.

Weave GameFabric Into Your Game.
Get Started
