Cloud Servers
Cloud servers are virtualized machines rented on demand from a provider rather than owned and racked. A hypervisor divides physical hardware among several tenants, so capacity can be added in minutes and released just as fast. You pay for what you run, and you share the underlying machine with workloads you cannot see.
Also called
Cloud hosting
,
How Is a Cloud Server Billed?
A cloud server is billed for the time it runs, by the hour, minute or second depending on the provider. Stop it and the compute charge usually stops too. Leave it running and it bills around the clock, with or without a match on it.
That is the trade against a bare metal server. The hourly rate is higher, and in exchange you can stop paying at any point. The saving only appears if something stops the server, and the provider won't. For game servers, that something is the orchestration layer: it starts a server when a match needs one and shuts it down when the match ends.
Compute is also only one line on the bill. Providers often charge separately for network egress per GB, and often for public IP addresses, storage and traffic between regions. Egress is the line that grows with players, because every player receives the match's state many times a second. All of these prices vary by region.
What Drives the Cost of Game Servers in the Cloud?
Three things, beyond the rate card:
Server size. Cloud servers come in fixed sizes. A game server sized for its worst case, or one match on an instance with room for four, pays for headroom it never uses. Load testing the build, and offering several sizes (small for a 1v1, larger for a full lobby), keeps each server close to what its match needs. See vCPU.
Shutdown. A server that finishes its match and stays up bills for nothing. So does a pool of servers kept running in case demand arrives.
Fill. A half-full match uses the same server time as a full one. How well the matchmaker fills sessions changes the cost per player directly.
None of the three is set by the provider. They are set by the server build, the matchmaker and the orchestrator.
What Do Cloud Servers Give a Multiplayer Game?
Elasticity and reach. Cloud capacity starts in minutes on a general-purpose provider, and faster on an orchestrator built for game servers, so a launch, a weekend peak or a content update is absorbed without ordering hardware. It also exists in far more places than a studio would rent machines, so a match can be placed near its players instead of at the nearest fixed host.
Capacity shared across many tenants, and across several providers, spreads the cost of each location. That is what lets a small game reach players on several continents without renting hosts on each.
Sharing a host does not mean sharing a game server's resources. Each server runs in a container with its own allocation of CPU and memory, so another tenant's load does not take them, and the noisy neighbour problem does not reach the match.
Games sit everywhere along the range. Fortnite runs "almost completely" on AWS, its game server fleet included (DatacenterDynamics, February 2023). Roblox builds and manages its own data centers for its backend and edge services, and uses public cloud "for burst capacity" (Roblox, January 2022). Most games land somewhere between the two.
On Edgegap, the Edge Cloud spans 615+ locations across 17+ providers (platform data, 18 September 2026), and each deployment goes to the best available location for its players at request time.
On Demand, Reserved or Spot: Which Cloud Pricing Fits Game Servers?
Cloud providers sell the same machine three ways.
On demand. Billed while it runs, at the full rate, with no commitment. It fits load that comes and goes: peaks, launches, regions with few players.
Reserved or committed. A lower rate in exchange for a term, typically one or three years. You pay whether servers run or not, the same trade as a bare metal host, at a cloud provider's price.
Spot. Spare capacity at a discount, which the provider can take back. On AWS, the warning is two minutes (AWS documentation, October 2026). A reclaimed server ends the match for every player on it, which makes spot capacity a poor fit for game servers.
A steady baseline is where committed capacity, in the cloud or on bare metal, usually earns its discount. On-demand capacity covers everything above it. Running both under one orchestrator is hybrid orchestration, and where the line sits differs for every game.
A word from our sponsor (ourselves!)
A region is a guess about where your players will be. Edgegap's Edge Cloud is a distributed, multi-cloud network of 615+ locations across 17+ providers, available on demand. Each server launches at the best available location on the network when the match starts, not at the nearest of a handful of regions.
Edgegap's Take (just our opinion, take it with a grain of salt!)
The Cloud Bill Is an Orchestration Bill
A cloud server bills until something stops it. What a game pays in the cloud depends less on the rate card than on two moments the provider does not handle: starting a server when a match needs one, and closing it when the match ends.
Starting fast is what makes starting per match possible. Our median deployment from a cold start is 2 seconds, over a rolling 30 days (platform data, 18 September 2026), so a server can be launched once players are matched instead of held running in case they are.
Closing is the half most often left to chance. On Edgegap, a game server can stop itself with one API call the moment its match ends, and each version can carry a maximum match duration, so a server that hangs stops on schedule instead of billing on (deployments documentation, October 2026). Billing is per minute, so the bill follows the matches.
Cloud and bare metal both belong in most games' infrastructure, and the mix is the studio's call. Paying only for the minutes each one is used is the orchestrator's job. To see what that mix costs at your own concurrency, model it in the pricing calculator.
,
