// MMO & MULTIPLAYER CO-DEVELOPMENT

The co-dev partner you call when the multiplayer problem gets hard.

We build and operate the parts of MMO and ambitious-multiplayer tech that most co-dev shops won't touch — dedicated servers, replication at scale, server-authoritative simulation, live ops. Embedded in your pipeline in 72 hours, shipping under your brand.

UE5 IRIS REPLICATION
NETOBJECTGROUPS
DELTA COMPRESSION
MOVER 2.0
SERVER-AUTH SIMULATION
GPU SKINNING
NANITE @ 70 CCU
INTEREST MANAGEMENT
LINUX DEDICATED SERVERS
TERRAFORM-MANAGED FLEET
UE5 IRIS REPLICATION
NETOBJECTGROUPS
DELTA COMPRESSION
MOVER 2.0
SERVER-AUTH SIMULATION
GPU SKINNING
NANITE @ 70 CCU
INTEREST MANAGEMENT
LINUX DEDICATED SERVERS
TERRAFORM-MANAGED FLEET
CASE STUDYATONETYPEMMORPGENGINEUnreal Engine 5DURATION5 YearsSTATUSOn hold

A multi-year MMO build, in the engine discipline we run on every project. Not a tech demo — a real one.

ATONE is the MMORPG we built — five years of full-cycle production for an external client, end-to-end across every discipline. The work that earned us the right to talk about MMO co-development. Five years of custom server architecture, replication tuning, and simulation work. Every problem a partner studio brings us, we've probably hit a version of it ourselves.

// 01
THE BRIEF

Dense, handcrafted layers at 70 CCU with full simulation.

ATONE's target wasn't open-world scale. It was the opposite: dense, art-directed layers where 70 CCU share a single simulation — full skeletal animation, physics-driven movement, networked ability effects, handcrafted Nanite geometry, and AI entities in the same frame.

70 CCU per layer is modest compared to a 2005 MMO. But when every character runs GPU skinning, Mover 2.0 physics, and IRIS-replicated state, you hit ceilings that a low-fidelity MMO never touches. When AI density broke our GAS-tick budget, we batched regular-mob ticks into one global update; the bottleneck shifted onto navmesh and perception — where we wanted it. That's the problem we signed up for, and the kind of instinct we now bring to partner projects.

// 02
WHAT WE BUILT

Custom dedicated server, UE5 IRIS replication stack, server-authoritative everything.

The stack is deliberately conservative on boundaries and aggressive inside them: all gameplay state is server-authoritative, clients predict and rewind, and replication is tuned per-feature rather than generic.

// ATONE — SIMPLIFIED ARCHITECTURE
┌─────────────────────────────────────────────────────────┐
│  CLIENT  ·  UE5  ·  Prediction + Rewind  ·  GPU Skin  │
└──────────────────────────┬──────────────────────────────┘
                           │  IRIS Replication
                           │  NetObjectGroups (area-filtered)
                           │  Delta codec (Mover 2.0 state)
                           ▼
┌─────────────────────────────────────────────────────────┐
│  DEDICATED SERVER  ·  Linux  ·  UE5 Standalone          │
│  Server-auth simulation · Interest mgmt · Anti-cheat │
└──────────────────────────┬──────────────────────────────┘
                           │  gRPC / Protobuf
                           ▼
┌─────────────────────────────────────────────────────────┐
│  BACKEND SERVICES  ·  Go  ·  PostgreSQL  ·  Redis         │
│  Auth · Characters · Inventory · Telemetry       │
└──────────────────────────┬──────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────┐
│  OPS  ·  Terraform  ·  K8s  ·  Grafana + Loki           │
│  Per-area shards · Blue/green deploys · Runbooks   │
└─────────────────────────────────────────────────────────┘

The dedicated server is a standalone UE5 build running Linux in containers. The replication layer uses IRIS with custom NetObjectGroups for area-based filtering — the first thing that falls over at 70 CCU without intentional scoping. Backend services are Go on PostgreSQL, with Redis for session and presence. The fleet is Terraform-managed with per-area shards and observability through Grafana plus Loki.

// 03
SHARP EDGES

Things that didn't work the first time. Things a partner studio can skip.

The honest value of a partner who's shipped the thing isn't that we know the happy path — it's that we know the ways it goes wrong. Here are four we can save you:

/01
IRIS is fast, but debugging what isn't replicating is harder.The legacy system had mature tooling; IRIS is newer. When an object stopped replicating in testing, our first debugging sessions were painful. We built a replication tracer that hooks IRIS's internal state and surfaces which NetObjectGroups a given object actually belongs to at runtime. We bring that tracer to client projects on IRIS.
/02
Nanite is not free in multiplayer.Prototypes looked great at 10 players. At 70 CCU with overlapping Nanite instances, GPU cost scaled in ways that single-player profiling didn't catch. Overlap resolution and instance culling needed rework. Budget these early in a networked Nanite project — not after the art pipeline is already flowing.
/03
Mover 2.0 state sync across 70 players: naive implementations break.Mover replicates a lot per player. Full state at 70 CCU at 20 Hz punched through our replication budget. We wrote a delta codec that ships only changed components, with a periodic full sync as anchor. Saved ~60% on movement bandwidth.
/04
GPU skinning interacts badly with cascaded shadows.The skinning pipeline was clean. The shadow pass at 70 characters wasn't. Cascaded shadow decisions have to be made with character density in mind from the start — switching strategies mid-project costs weeks of art and perf work.
// 04
WHAT IT MEANS
FOR YOUR
PROJECT

We come in with answers — or at least with the right questions.

The point of ATONE isn't that the MMORPG we built is the biggest or the fastest or the most successful. The point is that we've spent five years inside a production MMORPG making the same kinds of decisions you're making now — and we can come in on day three and tell you what we'd do and why.

The honest value of a partner who's shipped it isn't the happy path. It's knowing the ways it breaks — and what to do when it does.

Whether you're standing up a new server stack, inheriting one that's buckling under load, moving to IRIS, hitting a replication ceiling, or trying to get simulation cost under control at higher CCU — we've likely been there.

CCU per
area
70LIVE
Years in
production
5YRS
Movement
bandwidth saved
60%
Pipeline
integration
72HRS

See the ATONE visual case study →

What we actually do for you.

01

Dedicated Server Architecture

Design, stand up, and harden server-authoritative multiplayer stacks. UE5 standalone server builds, containerization, Linux deployment, interest management, anti-cheat integration. New stacks or rehabilitating inherited ones.

UE5 · UnityLinuxServer-auth
02

Replication & Netcode

UE5 IRIS migration, legacy replication tuning, custom NetObjectGroups, delta codecs for movement and ability state, prediction/rewind systems, RTT-aware behavior. We bring the tooling we built for ATONE — including the IRIS replication tracer.

IRISLegacy RepMover 2.0
03

Server-Authoritative Simulation

Gameplay simulation that holds up under load. AI behavior, ability systems, combat resolution, physics-driven movement, environment interactions — all authoritative on the server, with client prediction tuned to game feel.

GASAIPrediction
04

Persistence & Backend Services

Auth, character persistence, inventory, economy, matchmaking, telemetry, and the service mesh that ties them together. Go or C# backends on PostgreSQL, Redis, or your preferred stack. Observability as a first-class deliverable, not an afterthought.

Go · C#PostgreSQLgRPC
05

Ops & Infrastructure

Terraform-managed fleets, Kubernetes deployments, blue/green releases, per-area shard topology, capacity planning, runbooks your on-call team can actually use. Multi-region, low-latency matchmaking, DDoS hardening.

TerraformK8sGrafana
06

Live Operations

Post-launch feature development, content cadence, server ops, performance work, outage response, and the unglamorous engineering that keeps a multiplayer game healthy. We staff dedicated live-ops squads that carry your game for the long haul.

Feature DevContentOn-call

Three ways to bring us in. All of them start in 72 hours.

// START

Tell us where you're stuck.

Send your stack, your target scale, and the problem that's blocking you. We'll come back with a concrete read and a proposal — not a sales deck.

Send your problem
or grab the capabilities deck