Bare metal servers now cost 33% less on AccelByte Multiplayer Servers (AMS)

Starting Oct 1, 2026, bare metal instances on AccelByte Multiplayer Servers (AMS) cost 33% less in every region we offer them. Same machines, same regions, same fleets but a smaller bill. And bandwidth is included. This comes out of work on our own infrastructure. We've been steadily reducing the overhead in how we run bare metal capacity, and rather than keep what that saved, we're passing it through.

What's Changing

Bare metal on AMS is billed per instance-month. Here are the new rates:

cpx1.ser.16x — 16 vCPU, 32 GiB. For CPU-heavy simulation: physics, high tick rates.

glx1.ser.16x — 16 vCPU, 64 GiB. For general-purpose match servers, or packing more sessions onto one machine.

Prices are per instance-month in USD. Each instance runs on an Intel Xeon E-2288G or E-2388G, assigned at provisioning; the rate is the same either way. Current rates always live on the AMS pricing page.

Existing customers get the new rates automatically. There's nothing to migrate, re-provision, or request — your next invoice reflects the lower price.

Bare metal runs in the four regions above. Everywhere else — and for spikes anywhere — hourly cloud VMs cover nine regions across AWS, Azure and Google Cloud.

Bandwidth Is Included

This is the part that's easy to miss, and it's often worth more than the machine.

Every bare metal instance comes with outbound data transfer included at no charge:

Inbound traffic is never charged, on bare metal or anywhere else on AMS. Above the included allowance, outbound transfer is billed at our standard data transfer rates. Allowances are per region and aren't shared between them.

On hourly cloud VMs, that much egress in North America runs about $1,800 a month at our published rates — roughly eight times the $233 monthly price of the glx1 machine it's attached to. Egress is the line item that quietly decides the economics of a multiplayer game, and on bare metal, for most fleets, it simply isn't a line item.

Commit Longer, Pay Less

If your baseline is predictable, you can go further than the list price. Committing capacity for a fixed term takes another slice off the top:

These stack on the new rates, and the percentages are the same in every region. A glx1.ser.16x in North America on a two-year commitment works out to $198 a month — 43% below what the same machine cost yesterday.

The trade is straightforward: you're telling us how long you need the capacity, and we're pricing in the certainty. For a live game with a steady floor, that floor is the easiest thing in your infrastructure to commit to.

Already running bare metal? Raise a support ticket with the term you want and which instances it should cover, and we'll lock the rate in for that period.

Next to a Cloud VM

Bare metal and cloud VMs come in matching sizes on AMS, so this is a clean comparison rather than an apples-to-oranges one. A glx1.16x cloud VM is the same 16 vCPU and 64 GiB as a glx1.ser.16x bare metal machine — one is billed hourly, the other monthly.

Running that VM continuously for a month (730 hours) against the new bare metal price:

cpx1 lands in the same place. On a two-year commitment, it's 67% less. And all of that is before the bandwidth, which the VM pays for and bare metal doesn't.

There's a second difference that doesn't show up on the invoice: there's no hypervisor. A bare metal instance is a whole physical machine. Nothing sits between your server and the silicon, no other tenant is competing for your cores, and the cores themselves clock higher than the server parts sitting behind a comparable VM — which is the direction that matters for a simulation loop.

Some of that shows up as throughput. The part studios notice is consistency: tick times stop wobbling because of something happening on a host you can't see.

None of which is an argument for running everything on bare metal. A VM you switch off after a launch weekend is cheaper than a machine you rent for a month and use for two days — that's exactly what hourly capacity is for. The point is the floor: the capacity that never turns off is the capacity you should stop paying hourly rates for.

What It Means for Your Game

Most live multiplayer games have a floor and a ceiling. There's a baseline of players who are always on, and then there are spikes: a launch, a patch, a weekend, a tournament. Bare metal is the cheapest way to cover the floor, because you're buying a known amount of capacity at a known monthly price instead of paying by the hour for something that never turns off.

Taking a third off that monthly rate moves the math in a few useful directions.

More of your baseline belongs on bare metal. The crossover point where hourly cloud capacity becomes the cheaper choice now sits meaningfully higher, so a larger share of your steady traffic can run on fixed-price hardware. A studio holding 20 glx1 instances in Europe Central keeps $2,800 a month — $33,600 a year — without touching its architecture. On a two-year commitment it's $43,644.

Older games stay viable longer. A title past its launch window doesn't stop needing servers; it just stops earning what it used to. Cheaper baseline capacity pushes back the day the hosting bill forces a hard conversation about sunsetting.

Headroom stops being expensive. Most studios run their floor tighter than they'd like, because every spare machine is one you're paying for and not using. Buffer capacity for a patch day or an unexpected weekend now costs a third less to hold — and the cheapest way to avoid a queue is to already have the servers.