Hathora Shutdown 2026: Migration Alternatives After May 5

Hathora Cloud is no longer a future shutdown story. Hathora announced on March 4, 2026 that Fireworks AI had acquired the company, that the team would move to AI-inference orchestration, and that support for gaming customers would continue through May 5, 2026. Hathora also announced a migration partnership with Nitrado's GameFabric.

For teams choosing or re-evaluating a replacement now, the useful question is not “which vendor looks most like Hathora?” It is which parts of Hathora you were actually depending on: container execution, fleet allocation, regional capacity, server discovery, logs/metrics, matchmaking integration, or persistent game-backend services that lived somewhere else.

What actually happened

Hathora's own announcement says the company was acquired by Fireworks AI and that its team would work on compute orchestration for AI inference at scale. It also says Hathora's platform had spanned 14 regions, two bare-metal providers and four clouds, and names Splitgate 2, Stormgate and Predecessor among the live titles it powered.

The announcement does not say that the platform was frozen on acquisition day, that all customers were automatically transferred, or that every named game followed the same migration. What it does say is concrete: gaming-customer support continued through May 5 and Hathora partnered with Nitrado's GameFabric for a migration path and hands-on transition support.

GameFabric separately described the collaboration as a structured transition for Hathora customers and emphasized a Docker-native workflow intended to fit containerized game servers. That makes GameFabric the official continuity path, not proof that it is automatically the best architecture or price for every former Hathora workload.

Stormgate shows why the dependency boundary matters

Stormgate is the clearest public example of the operational blast radius. Its May 1, 2026 update disabled live-service Versus and Co-op, added offline support to Campaign, AI Skirmish and Custom modes, removed dependencies on Hathora and Pragma, and said its Steam peer-to-peer replacement was not yet ready.

That is a stronger lesson than “pick a safer vendor.” If core game modes only function while one orchestration path exists, the migration plan needs to include degraded-mode behavior: offline play where possible, a second fleet path, or at minimum a clean interface between matchmaking/session allocation and the game client.

Migration options in September 2026

GameFabric by Nitrado: the official transition path

GameFabric is the only option here with an officially announced Hathora transition partnership. Start here if preserving a Docker/container workflow and getting migration assistance matters more than redesigning the stack. Validate your actual requirements: regions, startup latency, allocation API, logs, autoscaling behavior, support and pricing, rather than assuming “official migration path” means exact feature parity.

Edgegap: usage-based global edge deployments

Edgegap's current public rate card lists on-demand Edge Cloud at $0.00115 per vCPU-minute with $0.10/GB egress, deploying across 615+ locations, plus reserved Private Fleet capacity. This model is attractive when you want broad placement and scale-to-zero economics, but network transfer is a real line item on the on-demand tier. Model your own session duration, CPU fraction and egress rather than comparing only the headline vCPU price.

Gameye: per-vCPU pricing with egress included

Gameye's pricing page, updated August 19, 2026, lists on-demand capacity from $0.07/vCPU-hour at a 1:2 vCPU/RAM ratio and reserved capacity from $0.027/vCPU-hour, with egress included and per-second billing. It also advertises Bring Your Own Infrastructure for teams that want Gameye's orchestration over their own bare metal/cloud accounts.

Amazon GameLift Servers: AWS-managed fleets

GameLift remains the obvious candidate when the rest of your operations already live in AWS. AWS currently offers managed EC2/container fleets, Spot and On-Demand capacity, autoscaling and multiple pricing models; eligible GameLift instance types now include network bandwidth at no extra charge. The trade-off is not simply “expensive AWS”: it is accepting AWS's fleet, IAM, observability and deployment model in exchange for a mature managed platform.

Crux: persistent backend first; Runtime is not a drop-in Hathora replacement

If your migration problem is auth, player data, leaderboards, economy, server registry or live config, Crux can own that persistent boundary independently of whichever fleet provider runs your game processes. That is the portable architecture this site is built around.

Crux Runtime itself should not be presented as a general Hathora replacement today. It is a capacity-gated closed alpha for Godot 4 Linux servers, currently one EU region, one fixed server shape and ephemeral 2-16 player sessions. If those constraints do not match your game, use a production fleet provider and keep Crux, if useful, on the persistent backend side.

A migration checklist that survives the next provider change

1. Inventory the Hathora contract you used. Container image/build upload, allocation API, regions, ports, readiness, logs, metrics, secrets, matchmaking callbacks and server discovery are separate dependencies.
2. Keep the game binary portable. Put provider-specific lifecycle glue behind a thin adapter rather than scattering one vendor's SDK throughout simulation code.
3. Separate durable state from match processes. A terminated session should not be the only copy of inventory, progression or purchase state.
4. Test allocation failure, not just happy-path launch. Exercise full regions, slow starts, failed health checks, capacity exhaustion and a provider API outage.
5. Run both paths before cutover. Canary real sessions on the replacement, compare startup/latency/error telemetry, then increase traffic while the old path is still available.
6. Keep an exit artifact. Record build format, environment variables, port contract, health/readiness semantics and allocation API mapping so the next migration starts from documentation rather than archaeology.

What changed since the original shutdown coverage

The shutdown date is now historical, so advice like “if you're still scrambling” is stale. Public pricing also moved: Gameye's current reserved rate is $0.027/vCPU-hour, not the older $0.025 figure. AWS GameLift now documents free network bandwidth for eligible instance types. The more durable takeaway is architectural: do not couple the fleet provider, persistent backend, matchmaker and player client more tightly than the game actually requires.

Related reading

• Unity Multiplay migration: Unity ended direct Multiplay hosting support March 31, 2026 and licensed the hosting software to Rocket Science Group.
• Game backend infrastructure stack: separate runtime, durable data and operational services before choosing vendors.
• AWS GameLift vs Agones: managed AWS fleets versus Kubernetes-owned orchestration.
• Backend cost calculator: model the persistent-service side separately from game-server compute.