Two game server hosting companies went dark within 60 days of each other in 2026. Unity shut down direct support for its Multiplay Game Server Hosting service on March 31, then Hathora followed on May 5 after announcing the closure a month earlier. For studios that had built their multiplayer backend around either platform, the shutdowns forced a decision almost nobody wanted to make mid-production: which cloud provider gets the migration.
The survivors left standing are a mixed group. Gameye, a Rotterdam-based orchestration platform running since 2017, publishes a flat $0.07 per vCPU-hour rate with no egress charges. Edgegap leans on edge distribution, claiming 615+ locations across 17+ infrastructure providers at roughly $0.069 per vCPU-hour plus usage-based bandwidth. AWS GameLift remains the default choice for studios already committed to Amazon’s cloud, now with bandwidth included free on most current fleets. This comparison pulls apart what each platform actually charges, how fast each one spins up a new server, what you have to build into your game binary to use them, and which one fits a given studio’s situation in October 2026.
None of this is hypothetical anymore. Teams that spent 2024 and 2025 building against Hathora’s SDK or Unity’s Multiplay REST API now have a hard deadline behind them, not in front of them. Hathora’s own migration partner page confirms the shutdown date and the forced timeline, and Unity’s status board shows the exact day new server allocations stopped working. The rest of this piece treats game server hosting providers as a live market with two fewer entrants than it had in January 2026, and ranks the remaining options on the numbers that show up on an infrastructure invoice rather than on marketing copy.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Timeline: How Hathora and Unity Multiplay Shut Down Within 60 Days
The sequence matters because it explains why so many studios are evaluating replacement platforms at the same time, which is itself driving up demand, and in some cases sales-call wait times, at the surviving vendors.
Unity announced first, telling Multiplay customers it was winding down direct support for the Multiplay Game Server Hosting Service. According to Unity’s own status notice, direct support ran through March 31, 2026, and as of April 1 customers could no longer scale new game servers or make new allocations on the Unity-run service. To avoid stranding live titles, Unity licensed the Multiplay codebase to Rocket Science Group, a company run by engineers who originally built the platform. Rocket Science now operates the same hosting technology independently, branded as Multiplay by Rocket Science, and its own site describes hybrid bare-metal and cloud hosting with automatic scaling across 31+ regions, per Multiplay’s current homepage.
Hathora moved next. The company announced on March 4, 2026 that it was discontinuing its game-hosting service, with the platform frozen and fully shut down within 90 days. That shutdown landed on May 5, 2026. Hathora’s own announcement stated the team had been acquired by Fireworks AI, an AI infrastructure company, marking a pivot away from game servers entirely. For the migration itself, Hathora partnered with Nitrado’s GameFabric platform, which now runs a dedicated Hathora migration page walking former customers through moving their Docker-based fleets onto GameFabric’s Agones-based infrastructure.
Both platforms being container-based turned out to matter a lot for studios on the way out. Game server binaries packaged as Docker images carry over to most other orchestration platforms without a rewrite, so the pain of migration landed mostly on configuration, matchmaker rewiring, and SDK replacement rather than a full rebuild of the server binary itself.
Why the Game Server Hosting Market Consolidated in 2026
Losing two platforms in the same quarter isn’t random noise in a niche market. Game server hosting sits in an awkward spot for venture-backed infrastructure companies: margins are thinner than general-purpose cloud because customers expect commodity-style pricing, yet the engineering lift to run a global, low-latency, DDoS-resistant network is just as heavy as any other infrastructure business. A market research report on the game server hosting platform category estimated that commodity cloud offerings from providers like AWS and Microsoft typically run $0.05 to $0.15 per player-hour, while specialized managed platforms range from $0.25 to $0.50 per player-hour, a spread that leaves little room for a smaller vendor to out-price a hyperscaler and still fund a growing sales and infrastructure team.
Hathora’s exit fits a pattern that’s become familiar across infrastructure startups in 2026: the team didn’t shut down and disappear, it pivoted into AI infrastructure. Fireworks AI, which builds inference infrastructure for AI models, absorbing the Hathora team suggests the underlying skill set, running low-latency, globally distributed compute at scale, transfers more easily into AI inference hosting right now than it does into continuing to fight for margin in game servers specifically. Unity’s move was different in kind: Multiplay wasn’t shut down so much as spun out, with Unity choosing to license the software to a team of its original engineers rather than keep running it as a line of business. Neither exit was a case of the underlying technology failing. Both were a case of the economics no longer justifying who was running it.
For studios, the practical lesson isn’t “avoid smaller vendors.” Gameye and Edgegap are themselves smaller, independent companies facing the same market pressure that pushed Hathora out. The more useful lesson is to weight published pricing, SLA commitments in writing, and a track record of continuous operation (Gameye’s 2017 founding date and Edgegap’s multi-year usage figures both predate this shutdown wave) more heavily than feature lists when picking a platform meant to run under a live game for years.
Gameye, Edgegap, and AWS GameLift at a Glance
Each of the three remaining platforms in this comparison takes a different bet on what multiplayer studios care about most.
Gameye bets that studios want a published number they can model against before a launch, not a sales call. It runs across 21 infrastructure providers and more than 200 datacenters, scales a new container in roughly 0.5 seconds, and bundles bandwidth into its flat per-vCPU rate rather than billing it separately. The company says it has orchestrated more than 130 million game sessions since founding in 2017 in Rotterdam, Netherlands, including the launch of Chivalry 2 at 250,000 concurrent players. Worth flagging directly: Gameye runs a competing orchestration platform, so its own comparison pages, which are among the more detailed public sources on this market right now, are not neutral, even where the underlying numbers check out against each vendor’s own pricing pages.
Edgegap bets on reach. Its 615+ locations across 17+ underlying infrastructure providers give it a genuinely distributed footprint, useful for titles where players are spread across regions a single cloud doesn’t cover well. Per Edgegap’s own published pricing page, the rate is usage-based at $0.00115 per vCPU-minute, about $0.069 per vCPU-hour, with a true $0 charge when no deployment is active, plus $0.10 per GB of monthly egress.
AWS GameLift bets on depth of ecosystem. If a studio already runs its backend, matchmaking, and analytics on AWS, GameLift slots into FlexMatch for rule-based matchmaking and FleetIQ for Spot-instance savings without adding a second cloud vendor to manage. The tradeoff is the GameLift Server SDK requirement in every build and AWS-only deployment with no cross-provider failover. For teams weighing this path, our Amazon GameLift server setup guide walks through the actual configuration steps.
Game Server Hosting Providers Compared: Full Specs Table
The table below pulls each row from the vendors’ own published documentation and pricing pages as of early October 2026. Where a figure is a vendor claim rather than an independently audited number, that is noted in the surrounding text rather than presented as settled fact.
| Criteria | Gameye | Edgegap | AWS GameLift |
|---|---|---|---|
| Published pricing | $0.07/vCPU-hr flat | $0.00115/vCPU-min (~$0.069/hr) | $0.741/hr for c6a.4xlarge (16 vCPU) = ~$0.046/vCPU-hr |
| Egress/bandwidth fees | None, included in rate | $0.10/GB monthly, $0 when idle | $0 on current eligible fleets per AWS’s bandwidth program |
| Free tier | Not published | Self-serve onboarding tier available | Free-tier eligible usage via AWS account, limited |
| SDK required in game binary | None, Docker + REST API | Optional | GameLift Server SDK, mandatory |
| Container scaling speed | ~0.5 seconds | Sub-second | Minutes for new EC2 capacity (cold start) |
| Infrastructure footprint | 21 providers, 200+ datacenters | 615+ locations, 17+ providers | AWS regions only, no cross-cloud |
| DDoS protection | Game-aware, applied at every provider | Included, traffic redirect and dispersal | AWS Shield Standard |
| Published SLA | 99.99% | Not formally published (99.99% availability commitment cited in vendor material) | 99.9% |
| DevOps overhead | Minimal, platform-managed | Minimal, self-serve | Moderate, fleet and SDK management |
| Matchmaker integrations | Pragma Engine, Nakama, Unity Matchmaker, FlexMatch, PlayFab | Platform-native matchmaking supported | FlexMatch (native), third-party via SDK |
| Container orchestration | Docker container + REST API, no Kubernetes exposure | Managed compute and managed Kubernetes | GameLift Fleets, optional containers |
| Cross-provider failover | Yes, multi-provider by design | Yes, multi-provider edge network | No, AWS-only |
| Track record cited | 130M+ sessions since 2017 | 14M CCU tested, 110M+ sessions lifetime | Years of production use across major titles |
Two rows deserve a second look before anyone treats them as a tiebreaker. The SLA figures come from each company’s own published commitments, not a third-party audit, and “DevOps overhead” is inherently a judgment call that depends on whether a studio already has a platform team on staff. Teams that want to run this stack themselves on Kubernetes rather than hand it to a managed vendor should look at our guide to running dedicated game servers on Kubernetes with Agones, which covers the DIY end of this same spectrum.
Pricing Breakdown: What You Actually Pay Per vCPU and Per GB
Raw hourly rates only tell part of the story, because egress, idle time, and minimum commitments change the real bill. Here’s what a studio running a mid-size fleet would actually see on an invoice.
Start with a simple, disclosed scenario: 100 dedicated server instances at 2 vCPUs each, running continuously for 720 hours, or 30 days. That works out to 144,000 vCPU-hours of compute, a number chosen purely for clean math, not pulled from any vendor’s own case study.
| Provider | Compute rate | Monthly compute (144,000 vCPU-hrs) | Egress (5TB/month example) | Estimated monthly total |
|---|---|---|---|---|
| Gameye | $0.07/vCPU-hr | $10,080 | $0 (included) | ~$10,080 |
| Edgegap | $0.069/vCPU-hr | $9,936 | $500 (5,000GB x $0.10) | ~$10,436 |
| AWS GameLift (c6a.4xlarge) | $0.046/vCPU-hr | $6,624 | $0 on eligible current-gen fleets | ~$6,624 + SDK integration cost |
On raw compute, GameLift comes out cheapest in this scenario, which cuts against the narrative that managed orchestration platforms always undercut the hyperscaler on price. AWS’s own GameLift pricing page backs this up with real examples: a 1v1 card game fleet sized for 10,000 peak concurrent users on c6a.large instances runs $3,237 per month in compute with $0 bandwidth charged, while a 100-player battle royale title on c6i.4xlarge instances for 100,000 peak CCU runs $165,217 per month, also with free bandwidth under AWS’s current program. Those figures come directly from AWS’s published calculator outputs, not an estimate.
What the GameLift number doesn’t capture is engineering time. Every one of those fleets requires the GameLift Server SDK wired into the game binary, ongoing fleet and FleetIQ management, and acceptance that failover only exists within AWS. Gameye and Edgegap both price in the convenience of skipping that SDK work, Gameye requires none at all and Edgegap’s is optional, and Edgegap’s willingness to charge $0 when a deployment is idle matters a lot more for bursty, tournament-style traffic than for a fleet running at constant utilization like the scenario above.
Scale the same math up to a launch-day fleet and the gap compounds. Using AWS’s own published example of a 100-player battle royale title on c6i.4xlarge instances sized for 100,000 peak concurrent users, AWS lists compute at $165,217 per month with $0 bandwidth charged. Applying Gameye’s flat $0.07 per vCPU-hour rate to the same instance count and utilization would land in a similar range before accounting for the fact that Gameye’s rate already includes bandwidth that AWS only recently made free, and applying Edgegap’s usage-based model would depend heavily on how bursty that 100,000-peak traffic actually is, since Edgegap’s $0 idle charge rewards exactly the kind of spiky, tournament-weekend demand pattern that a battle royale title tends to produce. The practical takeaway: at genuinely massive, sustained scale, the three platforms converge closer together than the smaller 100-instance scenario above suggests, because fixed costs like SDK integration and DevOps staffing get amortized across a much larger player base.
Benchmark Data: Scaling Speed, Cold Starts, and Uptime
Three separate sources give three different pictures of how fast these platforms actually respond to demand, and the gap between them is large enough to change a launch-day plan.
Gameye’s own published comparison claims containers scale in roughly 0.5 seconds across its 21-provider network, backed by a 99.99% published SLA. Edgegap advertises sub-second deployment with no pre-provisioning required and says it has been tested to 14 million concurrent users with more than 110 million lifetime sessions, plus a reported 100% measured uptime over the trailing 90 days as of September 2026 on its own comparison page. AWS GameLift, by contrast, provisions new EC2 capacity in minutes rather than seconds when a fleet needs to scale beyond its warm pool, a tradeoff of running on general-purpose cloud infrastructure rather than a pre-warmed container fleet. AWS’s documented SLA for GameLift sits at 99.9%, a full order of magnitude looser than the 99.99% figures the other two publish.
That gap matters most in the first hour after a launch, when demand is unpredictable and every minute of cold-start latency is a player watching a loading screen instead of playing. For steady-state workloads where traffic is forecastable days in advance, GameLift’s slower scaling is less costly because fleets can be pre-provisioned ahead of a known spike, such as a scheduled patch day or event.
None of these uptime and scaling figures come from an independent third-party benchmark lab. They’re self-reported by each vendor, and no public, apples-to-apples cold-start test across all three platforms exists as of this writing. Treat the SLA percentages as contractual commitments the vendor is willing to put in writing, not as measured outcomes from a neutral referee.
SDK and Matchmaker Requirements: What You Build Into Your Game Server
This is the row that decides how much engineering time a migration actually costs, and it is where the three platforms diverge the most.
Gameye requires no SDK at all. A studio ships a standard Docker container and calls a REST API to manage session lifecycle, which means the same container that ran on Hathora or Unity Multiplay can move over with configuration changes rather than a code change inside the game server binary. Edgegap’s SDK is optional, giving studios a middle path: skip it for basic deployment, add it if deeper lifecycle hooks are needed. GameLift sits at the other end, requiring the GameLift Server SDK compiled into every build, which means any studio moving to GameLift from a container-only platform is looking at actual code changes, not just configuration.
Engine and Matchmaker Integrations
Matchmaker support follows a similar pattern. Gameye lists native integrations with Pragma Engine’s official ServerProviderPlugin, Nakama’s Fleet Manager, Unity Matchmaker, AWS FlexMatch, and PlayFab, giving studios a config-level path to swap matchmakers without re-architecting. GameLift’s FlexMatch is a mature, rule-based matchmaker but it’s AWS-native, so pairing it with a non-AWS game server platform adds integration work. Studios already running backend services through Unity Gaming Services or PlayFab should also weigh how those platforms’ own multiplayer tooling compares, which our Unity Gaming Services vs PlayFab vs Beamable comparison covers in more depth.
DDoS Protection and Security Posture
All three platforms ship some form of DDoS mitigation, but the implementation details differ enough to matter for competitive titles where even a brief attack during a ranked match is a support ticket nightmare.
Gameye applies what it calls game-aware DDoS protection at every provider in its network, meaning the mitigation layer understands game traffic patterns specifically rather than applying generic web-traffic filtering. Edgegap’s platform detects abnormal traffic patterns in real time and automatically redirects traffic away from the targeted server while dispersing the malicious load, scaling resources during the attack itself rather than just absorbing it. AWS GameLift relies on AWS Shield Standard, the baseline DDoS protection bundled with AWS accounts, with AWS Shield Advanced available as a paid upgrade for studios that need the higher tier.
One nuance worth flagging for edge-distributed platforms generally: when a network spans many underlying infrastructure providers, as Edgegap’s 615+ locations do, hardware quality and the strength of DDoS protection can vary by the specific provider behind any given location. A curated, smaller network gives more predictable performance for latency-sensitive competitive play, even if it covers fewer total points on a map.
Infrastructure Footprint: Edge Reach vs Curated Datacenters vs AWS Regions
How a platform distributes its servers geographically ends up shaping which games should even consider it.
What 615+ Locations Actually Means
Edgegap’s headline number, 615+ locations across 17+ infrastructure providers, is genuinely the largest footprint among the three. But it is a count of locations across partner infrastructure, not 615 datacenters Edgegap itself built and operates. For a globally distributed player base, particularly in regions underserved by the big three clouds, that breadth can mean meaningfully lower latency for players in markets a curated network skips. Gameye’s 200+ datacenters across 21 providers takes a similar multi-provider approach at a smaller scale, while GameLift is limited to wherever AWS runs a region, which is extensive but not infinite, and offers zero coverage outside AWS’s own footprint with no cross-provider failover if an AWS region has an incident.
Studios that want to avoid vendor lock-in to any single cloud, including the orchestration layer itself, sometimes look at serverless or edge-compute alternatives instead of traditional dedicated server fleets entirely. Our Cloudflare Durable Objects game server setup guide covers one such approach for titles where the edge-network model fits better than a dedicated-fleet model.
Teams that want the self-hosted, full-control end of this spectrum rather than any managed vendor can also run their own Agones-based orchestration layer on a cloud Kubernetes service. Our guide to running Agones game servers on Oracle Cloud’s OKE shows what that setup actually involves, including the DevOps burden all three managed platforms in this comparison are designed to remove.
Nitrado GameFabric and Agones: The Kubernetes-Native Middle Ground
Gameye, Edgegap, and AWS GameLift aren’t the only names a studio researching this market will run into, and leaving out the fourth and fifth most-cited options would understate how crowded this decision actually is.
Nitrado’s GameFabric is the platform Hathora itself pointed customers toward during its shutdown, which makes it worth a closer look even though it sits outside this article’s main three-way comparison. GameFabric runs on Kubernetes, built on top of Agones, across Nitrado’s own network of 67+ locations, a vertically integrated approach that trades the breadth of Edgegap’s 615+ locations for a single operator’s consistency. The tradeoffs show up at the application layer: GameFabric requires Agones SDK lifecycle calls inside every build, the same requirement as running Agones directly, and as of this writing it has no documented native integration for Pragma Engine or Nakama the way Gameye does. Pricing isn’t published, so studios evaluating it need a sales conversation before they can model costs, the same friction that pushed some former Hathora customers toward Gameye’s published-rate model instead.
Agones itself, the open-source Kubernetes-native game server framework backed by Google and running underneath GameFabric, remains the right call for studios with a dedicated platform or DevOps function that wants to own the entire stack rather than hand it to any vendor. It’s free software, but “free” only covers the orchestration layer. The studio still pays its cloud provider for the underlying Kubernetes nodes and control plane, runs its own on-call rotation, and implements Agones SDK lifecycle calls in its own binary. Our Agones on Oracle Cloud OKE walkthrough and general Kubernetes and Agones setup guide both cover what that actually looks like in practice, including the cluster configuration work that a managed platform like Gameye or GameFabric is specifically selling studios out of having to do themselves.
Real-World Examples: Five Studios, Five Different Calculations
Pricing tables only go so far. Here’s how the decision plays out in five different situations, three drawn from published case material and two built as disclosed, illustrative scenarios rather than real named studios.
- Chivalry 2 (Torn Banner Studios), via Gameye: According to Gameye’s own case study material, the melee combat game launched at 250,000 concurrent players on Gameye’s infrastructure with zero downtime reported. This is a vendor-published claim rather than an independently audited figure, but it is the most specific launch-scale number publicly available for any platform in this comparison.
- Clone Drone in the Danger Zone (Doborog Games), via Gameye: Gameye’s published material states the indie studio cut server costs by more than 60% after switching to the platform, attributing most of the savings to eliminating egress charges entirely rather than to compute pricing alone.
- Orion Drift (Another Axiom), via Gameye: Listed by Gameye as a current customer in its published case material, though without the same level of specific performance detail as the Chivalry 2 example.
- Illustrative scenario, globally distributed indie co-op title: A small studio with players concentrated in Southeast Asia, Eastern Europe, and South America, regions a single hyperscaler region often serves with higher latency, would likely weigh Edgegap’s 615+ location footprint and self-serve onboarding most heavily, accepting usage-based egress billing as the cost of that reach.
- Illustrative scenario, AWS-native live-service shooter: A studio that already runs its backend services, analytics pipeline, and matchmaking on AWS would have the least friction sticking with GameLift and FlexMatch, especially given that AWS’s own pricing examples show bandwidth now running free on current-generation fleets, even accepting the SDK integration work and single-cloud exposure.
Who Should Pick Which Platform: Five Use Cases
Pulling the specs and pricing together into direct recommendations:
- Studios migrating off Hathora or Unity Multiplay on a deadline: Gameye’s no-SDK, Docker-container model is the lowest-friction landing spot, since existing container images transfer without a code change in the game server binary.
- Competitive and AAA titles needing predictable launch-day scale: Gameye’s published 99.99% SLA and sub-second scaling, or GameLift if the studio is already AWS-committed and willing to accept slower cold starts in exchange for ecosystem depth.
- Indie and globally distributed titles prioritizing geographic reach: Edgegap’s 615+ location footprint and self-serve, free-tier-friendly onboarding fit smaller teams without large DevOps budgets.
- Studios already standardized on Pragma Engine or Nakama for matchmaking: Gameye is currently the only platform in this comparison with documented native integrations for both, turning a migration into a configuration change rather than a rebuild.
- Teams with an in-house platform or DevOps function that want full infrastructure control: None of these three managed platforms is the right fit. Agones running on self-managed Kubernetes, as covered in our Kubernetes and Agones setup guide, gives up managed convenience in exchange for owning the entire stack.
Migration Guide: Moving Off Hathora or Unity Multiplay Without a Rewrite
For studios still sitting on a Hathora or Unity Multiplay deployment past its shutdown date, the practical migration path depends on how much of the original architecture is reusable.
The good news, confirmed by Gameye’s own technical material and consistent with how both shut-down platforms operated, is that both Hathora and Unity Multiplay ran on container-based deployment. That means the actual game server binary, packaged as a Docker image, typically transfers to another container-based platform without modification. What has to change is everything around the container: the orchestration API calls, the matchmaker integration, session lifecycle hooks, and any platform-specific SDK calls the game code was making directly.
Step-by-Step Migration Checklist
- Audit every direct API call to the old platform’s SDK inside the game server binary and flag each one for replacement or removal.
- Confirm the existing Docker image builds and runs standalone outside the old platform’s management layer, which validates the container itself is portable.
- Pick a destination platform based on SDK requirements: no-SDK platforms like Gameye require the least binary-level rework, while GameLift’s mandatory SDK means budgeting real engineering time.
- Reconnect the matchmaker. If the studio used Unity Matchmaker, Nakama, or Pragma Engine, check whether the destination platform has a native integration before writing custom glue code.
- Re-point session allocation and server discovery calls to the new platform’s API or REST endpoint.
- Run a load test at a fraction of expected launch-day concurrency before cutting DNS or client-side server discovery over to the new platform.
- Keep the old platform’s remaining uptime window, if any, as a rollback window rather than cutting over on the deadline itself.
Former Unity Multiplay customers have an additional option the Hathora exodus didn’t: Rocket Science Group’s licensed continuation of the original Multiplay codebase, branded Multiplay by Rocket Science. Per the company’s own site, this is explicitly positioned as the same underlying hosting technology, which makes it the lowest-effort continuation path for a live title, at the cost of moving to a newer, smaller, and less-established vendor than AWS or Gameye, and one that has not yet published its own pricing.
Sample code showing the general shape of a container-based session allocation call, the kind of integration point that has to be rewritten against a new platform’s API during migration:
curl -X POST https://api.example-platform.com/v1/sessions \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"image": "registry.example.com/mygame-server:latest",
"region": "us-east",
"cpu": 2,
"memory_mb": 2048,
"port_mappings": [{ "container_port": 7777, "protocol": "udp" }]
}'Every platform’s actual request shape differs, but the pattern of specifying a container image, region, resource limits, and port mapping is common enough across Gameye, Edgegap, and GameLift’s container mode that most of the migration work is substitution rather than invention.
Pros and Cons Side by Side
Each platform’s strengths map directly onto its weaknesses, which is normal for infrastructure choices with this many tradeoffs baked in.
| Platform | Pros | Cons |
|---|---|---|
| Gameye | Published flat pricing, no egress fees, no SDK required, sub-second scaling, broad matchmaker integrations | Not an edge-first network like Edgegap, managed orchestration means less low-level control than Agones |
| Edgegap | Largest stated footprint (615+ locations), self-serve onboarding, optional SDK, $0 cost when idle | Usage-based bandwidth can add up at scale, hardware quality varies by underlying provider, SLA not formally published |
| AWS GameLift | Cheapest raw compute rate in this comparison, free bandwidth on eligible fleets, deep AWS ecosystem (FlexMatch, FleetIQ), proven at large scale | Mandatory SDK in every build, AWS-only with no cross-provider failover, cold starts measured in minutes not seconds, lower 99.9% SLA |
The Verdict: What the Numbers Actually Say
There isn’t a single winner here, because the three platforms are optimized for different constraints rather than competing on the same axis.
On pure compute cost, AWS GameLift’s $0.046 per vCPU-hour on a c6a.4xlarge instance beats both Gameye’s $0.07 and Edgegap’s $0.069, and AWS’s current bandwidth program removes what used to be GameLift’s biggest hidden cost. That price advantage comes paired with the GameLift Server SDK requirement in every build and a hard AWS lock-in, costs that show up in engineering time rather than on the infrastructure invoice.
For studios that weight scaling speed and operational simplicity over raw compute price, Gameye’s published 0.5-second scaling and zero-SDK model make it the more straightforward migration target for anything coming off a container-based platform like Hathora or Unity Multiplay, a case its own marketing leans on heavily but one that holds up against its published technical claims. For studios with players scattered across regions a single cloud underserves, Edgegap’s 615+ location footprint is the most geographically complete option on this list, even with usage-based bandwidth billing attached.
The honest bottom line for October 2026: run the numbers against your own fleet size and player distribution before picking based on any single headline figure, including the ones in this article’s own pricing table. A 30% swing in average concurrent users relative to peak, which is common for live-service games, changes which platform wins on cost by enough to flip the recommendation.
Frequently Asked Questions
What happened to Hathora and Unity Multiplay in 2026?
Unity ended direct support for its Multiplay Game Server Hosting Service on March 31, 2026, licensing the codebase to Rocket Science Group. Hathora announced its shutdown on March 4, 2026 and fully closed on May 5, 2026, with the team moving to Fireworks AI and Nitrado’s GameFabric serving as the designated migration partner.
Which game server hosting provider is cheapest in 2026?
On raw compute pricing, AWS GameLift’s published rate of roughly $0.046 per vCPU-hour on a c6a.4xlarge instance undercuts both Gameye ($0.07/vCPU-hr) and Edgegap (~$0.069/vCPU-hr), especially now that bandwidth runs free on eligible current-generation fleets. That advantage narrows or disappears once a studio accounts for the engineering time needed to integrate the mandatory GameLift Server SDK.
Do any of these platforms require an SDK in my game server binary?
AWS GameLift requires the GameLift Server SDK in every build. Edgegap’s SDK is optional. Gameye requires no SDK at all, using a standard Docker container plus REST API calls for session lifecycle management.
Which platform charges egress or bandwidth fees?
Edgegap bills bandwidth at $0.10 per GB of monthly egress, with a $0 charge when no deployment is active. Gameye includes all bandwidth in its flat $0.07 per vCPU-hour rate with no separate egress charge. AWS GameLift currently offers free bandwidth on eligible fleets under its network bandwidth program, a meaningful recent change from its historical per-GB billing.
Is Rocket Science Group a good alternative to AWS GameLift or Gameye for former Unity Multiplay customers?
Rocket Science Group, which licensed the Multiplay codebase, offers the lowest-effort continuation for a title that was already running on Unity Multiplay, since it is the same underlying hosting technology. It is also a newer, smaller vendor still transitioning the platform onto its own infrastructure, and as of this writing it has not published its own pricing, which makes cost modeling harder than with GameLift or Gameye.
How fast do these platforms scale up a new game server?
Gameye publishes roughly 0.5-second container scaling. Edgegap advertises sub-second deployment with no pre-provisioning. AWS GameLift, which provisions new EC2 instances rather than lightweight containers when scaling beyond a warm pool, takes minutes for a cold start. None of these figures come from an independent, audited benchmark.
Can I move a Docker-based game server from Hathora or Unity Multiplay without rewriting it?
Generally yes, because both discontinued platforms ran container-based deployment. The game server binary inside the Docker image typically transfers without code changes. What requires rework is the surrounding integration: matchmaker calls, session lifecycle hooks, and any direct SDK calls the game was making to the old platform specifically.
Which platform has the best DDoS protection for competitive multiplayer games?
All three include some form of mitigation. Gameye describes its protection as game-aware and applied at every provider in its network. Edgegap’s system detects abnormal traffic and automatically redirects and disperses it while scaling resources during an attack. AWS GameLift relies on baseline AWS Shield Standard, with AWS Shield Advanced available as a paid upgrade for studios that need stronger guarantees.
What is Nitrado GameFabric and should I consider it instead of these three?
GameFabric is Nitrado’s Kubernetes-based orchestration platform, built on Agones and running across 67+ of Nitrado’s own locations. It was Hathora’s designated migration partner during its shutdown. It’s worth considering if a studio wants a single vertically integrated network and is comfortable with Agones SDK lifecycle calls, but it doesn’t publish pricing and has no documented Pragma Engine or Nakama integration, unlike Gameye.
Why did Hathora and Unity Multiplay shut down if the underlying technology worked?
Both exits appear to be about economics rather than product failure. Hathora’s team moved into AI infrastructure through its acquisition by Fireworks AI, a sector currently attracting far more capital than game server hosting. Unity chose to license Multiplay to Rocket Science Group rather than continue running it as a business line itself, which kept the technology alive under new ownership instead of shutting it down outright.
