Unity pulled the plug on its own Multiplay Game Server Hosting service on April 1, 2026, closing a chapter that once made Unity the default answer for studios that needed dedicated servers without a dev-ops team. The company stopped accepting new game-server allocations and told customers to either migrate to a new independent operator called Multiplay by Rocket Science or pick a different cloud platform entirely. For thousands of multiplayer studios, that single status-page notice turned a routine infrastructure decision into an urgent one.
The studios affected now face three realistic paths. They can follow the legacy Multiplay codebase to Rocket Science Group, the company founded by former Multiplay engineers that licensed the platform from Unity. They can move to Amazon’s purpose-built dedicated server product, AWS GameLift Servers. Or they can build their own orchestration layer on Kubernetes using Google’s open source Agones project. Each path carries a different cost structure, a different operational burden, and a different bet on who controls the roadmap. This comparison breaks down the pricing, the migration mechanics, and the trade-offs so a studio engineering lead can make the call before the next billing cycle closes.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Why Unity Multiplay Shut Down and What It Means for Game Server Hosting
According to Unity’s own status notice, the company gave customers a March 31, 2026 cutoff to declare their migration intent, then formally deprecated the Multiplay Game Server Hosting service the next day. Per Unity’s status page, customers can no longer scale new game servers or make new fleet allocations on the legacy service. The one carve-out: studios that clearly signaled before the deadline that they intended to move to Multiplay by Rocket Science were allowed to keep running on the platform until that migration finished.
That carve-out matters because it is the clearest signal of what Unity actually did here. Rather than simply killing the product, Unity licensed the Multiplay technology stack to an outside operator and stepped back from running game server hosting as a first-party business. Rocket Science Group, staffed by veterans of the original Multiplay engineering team, now runs the platform independently. In a company announcement, Rocket Science said it had already transitioned the business “several months” earlier and now supports more than 60 live clients, including what it described as a successful global, multiplatform launch.
For studios, the practical upshot is a forced evaluation. Game server hosting is a classic build-vs-buy decision, and Unity’s exit removed one of the “buy” options from the table in its original form. The replacement options split into three philosophies: stick with the familiar Multiplay API under new ownership, lean on a hyperscaler’s managed game-hosting product, or take full control with open source Kubernetes tooling. Industry data from market researchers tracking the game server hosting platform sector describes Kubernetes-based orchestration as already managing roughly 70% of cloud-native game server environments, with auto-scaling systems built to absorb traffic spikes of around 300% during major content releases. That context explains why Agones, not just AWS and the Multiplay successor, belongs in this comparison.
The Broader Game Server Hosting Market in 2026
The Unity Multiplay shutdown did not happen in a vacuum. It landed in the middle of a broader shift in how multiplayer games get hosted. Industry tracking from Business Research Insights, cited earlier, puts Kubernetes-based orchestration at roughly 70% of cloud-native game server environments already, which helps explain why Agones carries real weight in this comparison rather than being a niche afterthought next to two commercial products. That same research describes autoscaling systems in this space as routinely built to absorb traffic spikes of around 300% during major content releases, a swing that would overwhelm a statically provisioned fleet in minutes.
The same body of research reports that roughly 55% of hosting providers have rolled out AI-driven predictive scaling models specifically to get ahead of those surges rather than simply reacting to them after player counts spike. That statistic matters for this comparison because none of the three platforms here are static products. AWS continues to extend GameLift’s FleetIQ autoscaling, the Agones project ships a new minor version roughly every month with scheduling and health-detection improvements, and Rocket Science has positioned its relaunch explicitly around supporting “a successful global, multiplatform launch” as proof its scaling can handle a real live-service workload, not just a migrated legacy fleet coasting on existing infrastructure.
What this means practically for a studio weighing its options: the choice is not just about today’s pricing and today’s feature set. It’s about which platform’s roadmap and operational maturity you’re betting your next two or three years of live-service operations on. A platform that is still actively shipping fixes for health detection and rolling-update stability, as Agones was in its September 2026 release, is signaling continued investment. A platform whose roadmap depends on one company’s board-level strategy, as Unity Multiplay’s did, carries a structural risk that this entire article exists because of.
The Three Migration Paths at a Glance
Before diving into pricing and specs, it helps to frame what each platform actually is, because they are not three versions of the same product. AWS GameLift Servers is a fully managed, pay-per-instance dedicated server hosting service that AWS operates end to end. Google Cloud Agones is not a managed service at all — it is free, open source software that runs on top of a Kubernetes cluster you provision and pay for separately, most commonly on Google Kubernetes Engine (GKE). Multiplay by Rocket Science is the most unusual of the three: a commercially operated platform running the same underlying technology that Unity built, now maintained by an independent company with its own pricing and support relationship.
That distinction drives almost every other difference in this comparison. GameLift and Rocket Science are both “someone else operates the control plane” products. Agones is a “you operate everything, we give you the scheduling logic” product. A studio choosing between these isn’t really picking a vendor on price alone, it’s picking how much infrastructure engineering it wants to own.
Specs Comparison: GameLift vs Agones vs Multiplay by Rocket Science
| Attribute | AWS GameLift Servers | Google Cloud Agones | Multiplay by Rocket Science |
|---|---|---|---|
| Hosting model | Fully managed, AWS-operated fleets | Open source orchestration on self-managed Kubernetes/GKE | Fully managed, third-party operated platform |
| Underlying infrastructure | AWS EC2 instances (On-Demand or Spot) | Any Kubernetes cluster (GKE, EKS, AKS, on-prem) | Rocket Science-operated infrastructure (licensed Multiplay stack) |
| Pricing basis | Per-instance-hour, by instance family | No license fee; you pay for GKE/compute only | Not publicly listed; commercial contract |
| Free bandwidth | Free in/out for Gen-6+ instance types | Standard cloud provider egress pricing applies | Not publicly disclosed |
| Spot/preemptible support | Yes, roughly 50-85% cheaper than On-Demand | Yes, via Kubernetes preemptible node pools | Not publicly disclosed |
| Current software version | GameLift Servers (continuously updated AWS service) | Agones 1.61.0 (released September 2026) | Rocket Science-maintained Multiplay fork |
| Kubernetes compatibility | Not applicable (proprietary fleet model) | Kubernetes 1.34, 1.35, 1.36 | Not applicable |
| Autoscaling control | AWS-managed FleetIQ autoscaling | Self-configured Fleet Autoscaler | Platform-managed (vendor-controlled) |
| Operational ownership | Low, AWS runs the control plane | High, your team runs the cluster | Low, Rocket Science runs the platform |
| Migration path from Unity Multiplay | Manual re-integration via GameLift SDK | Manual re-integration via Agones SDK plus K8s setup | Direct, same underlying codebase as legacy Multiplay |
| Hybrid/bring-your-own-hardware option | Yes, via GameLift Anywhere | Yes, any Kubernetes cluster including on-prem | Described as hybrid hosting and orchestration |
| Vendor lock-in risk | Moderate (AWS-specific APIs) | Low (open source, portable across clouds) | Moderate to high (single commercial vendor) |
A few rows in that table deserve a closer look before any studio commits budget. The “current software version” row is not cosmetic. Per the Agones release notes, version 1.61.0 shipped in September 2026 with a hardened SDK sidecar security context, a fix for GameServer health-detection bugs, improved Fleet rolling-update stability, and a resolved memory leak. That release supports Kubernetes 1.34 through 1.36, according to the Agones installation documentation. Studios running older Agones builds on older Kubernetes versions should treat an upgrade as a prerequisite, not an afterthought, before using any of the cost figures below.
Pricing Breakdown: What Each Platform Actually Costs
Pricing is where the three paths diverge most sharply, and it’s also where the comparison gets genuinely complicated, because AWS and Google price compute by the hour while Rocket Science prices a relationship. Here is what is publicly documented as of October 2026.
| Instance/Resource Type | Provider | Published Rate | Notes |
|---|---|---|---|
| c4.large (2 vCPU, 3.75 GiB) | AWS GameLift Servers | $0.127/hour | On-Demand Linux pricing |
| c5.large (2 vCPU, 4 GiB) | AWS GameLift Servers | $0.109/hour | On-Demand Linux pricing |
| c5.xlarge (4 vCPU, 8 GiB) | AWS GameLift Servers | $0.211/hour | On-Demand Linux pricing |
| c5a.large (2 vCPU, 4 GiB) | AWS GameLift Servers | $0.099/hour | On-Demand Linux pricing |
| c5d.large (2 vCPU, 4 GiB) | AWS GameLift Servers | $0.122/hour | On-Demand Linux pricing |
| c6i.large (2 vCPU, 4 GiB) | AWS GameLift Servers | $0.109/hour | On-Demand Linux pricing |
| c6g.large (2 vCPU, 4 GiB, Graviton) | AWS GameLift Servers | $0.088/hour | Cheapest listed On-Demand rate; Graviton price-performance edge |
| GameLift Spot instances | AWS GameLift Servers | 50-85% below On-Demand | Requires session-recovery logic for interruptions |
| GKE cluster management fee | Google Cloud (Agones host) | $0.10/cluster-hour | Flat fee on top of node compute, applies to Standard and Autopilot |
| GKE Autopilot vCPU-hour | Google Cloud (Agones host) | $0.0133 – $0.0445/vCPU-hour | Varies by region; pods billed on requested CPU/memory |
| GKE free tier credit | Google Cloud (Agones host) | $74.40/month per billing account | Covers roughly one small Autopilot or zonal Standard cluster |
| Multiplay by Rocket Science | Rocket Science Group | Not publicly listed | Commercial quote required; 60+ live clients claimed |
The headline figure worth flagging is the gap between AWS’s cheapest listed GameLift instance type and Google’s cheapest Autopilot tier. At $0.088 per hour for a c6g.large Graviton instance versus roughly $0.0133 per vCPU-hour on GKE Autopilot’s lowest regional tier, the raw compute math can favor Agones-on-GKE for teams that are comfortable running their own Kubernetes stack, but that number ignores the GKE cluster management fee of $0.10 per cluster-hour that applies on top of node costs, plus the engineering time needed to build, test, and operate the Agones Fleet Autoscaler yourself. AWS, according to its own GameLift Servers pricing page, markets a blended scenario where combining free bandwidth on Gen-6+ instances, Spot capacity, and Graviton processors can bring effective hosting costs down to roughly $1 per player per month, though AWS is explicit that this is an illustrative optimization scenario, not a flat universal rate.
Rocket Science is the wildcard because its pricing simply is not public. That is consistent with how legacy Unity Multiplay pricing worked too: enterprise game hosting contracts have historically been negotiated per studio based on concurrent player counts, regions, and SLA requirements rather than published as a rate card. Any studio evaluating Rocket Science needs to request a quote directly; treat any third-party estimate of Rocket Science pricing as unverified until confirmed in a signed contract.
Benchmarks and Performance Signals
Hard, apples-to-apples cold-start and latency benchmarks across all three platforms are not publicly available as of October 2026, and studios should be skeptical of any comparison that claims otherwise without disclosing its exact test methodology, instance types, and regions. What is independently documented is more structural than a stopwatch test, but it still shapes real-world performance expectations.
- AWS’s free-bandwidth policy, confirmed on its GameLift instance pricing page, applies to both On-Demand and Spot capacity on Generation-6-and-newer instance types across supported regions, which removes a variable cost that otherwise scales unpredictably with concurrent player counts.
- Agones 1.61.0’s fix for GameServer health-detection issues, noted in the project’s own release notes, directly affects how quickly a fleet can tell the difference between a crashed server and a slow one, a distinction that drives both player experience and autoscaling accuracy.
- Market analysis from Business Research Insights describes roughly 55% of hosting providers as having adopted AI-driven predictive scaling models specifically to manage player-count surges that exceed normal traffic patterns, a trend that applies pressure on all three platforms to improve reaction time to demand spikes.
- GameLift Spot’s advertised 50-85% discount band versus On-Demand pricing is the single largest lever any of these three platforms offers for cutting raw compute spend, provided the game architecture can tolerate a server being reclaimed mid-session.
The practical lesson is that fast and cheap are not separable questions here. A Spot-heavy GameLift fleet or a preemptible-node-heavy Agones cluster can hit excellent cost-per-player numbers, but only if the game’s session-recovery and reconnect logic is built to handle interruptions gracefully. Teams that skip that engineering work and simply flip on Spot or preemptible capacity tend to discover the gap the hard way, during a launch weekend traffic spike.
Real-World Deployment Scenarios
Because Rocket Science has not published a client roster and neither AWS nor Google has released named case studies specific to post-Multiplay migrations as of this writing, the scenarios below are drawn from common, verifiable deployment patterns in the game server hosting space rather than attributed to specific unnamed studios. They illustrate how the decision plays out in practice.
- A mobile battle royale title with unpredictable regional surges. A game that spikes hard around content drops and esports events typically favors GameLift’s FleetIQ autoscaling combined with Spot capacity, since AWS’s own pricing guidance is built around exactly this bursty-demand pattern.
- A mid-size PC co-op shooter migrating off legacy Unity Multiplay. Studios with existing Multiplay SDK integrations and limited engineering bandwidth to rebuild server-fleet logic are the most natural fit for Multiplay by Rocket Science, since the underlying codebase and session APIs carry over directly.
- A platform-agnostic live-service game already running Kubernetes for its backend. Studios that already operate GKE or EKS clusters for matchmaking, inventory, or analytics services gain the most from Agones, because it slots into infrastructure and on-call processes the team already owns rather than introducing a second vendor relationship.
- An indie studio with one or two engineers and no dedicated infrastructure hire. The operational simplicity of a fully managed product outweighs marginal cost savings here, pointing toward GameLift or Rocket Science over a self-managed Agones cluster that needs patching, monitoring, and on-call coverage.
- A publisher running the same game across AWS, Azure, and on-prem data centers for regulatory reasons. Agones’s Kubernetes-native portability is the only one of the three options built to run identically across clouds and bare-metal hardware without a rewrite, which matters for studios facing data-residency requirements in specific markets.
AWS GameLift Servers: Strengths and Trade-offs
GameLift’s core appeal is that AWS operates the entire control plane, so a studio’s engineering team spends its time on gameplay code instead of fleet management. The free-bandwidth policy on Gen-6+ instances is a genuine cost advantage that is easy to underestimate, since network egress is one of the least predictable line items in any game server budget, scaling directly with concurrent player count and match duration. Removing that variable from eligible instance types turns a volatile cost into a fixed one.
GameLift also offers a hybrid option called GameLift Anywhere, which lets a studio register its own hardware, whether that’s on-prem servers, a colocation facility, or a different cloud, into the same fleet-management APIs AWS provides for EC2-based fleets. That matters for studios that want AWS’s orchestration tooling without being fully locked into AWS compute pricing for every region they operate in.
The trade-off is architectural lock-in. GameLift’s SDK, session-placement model, and fleet APIs are AWS-specific. A studio that builds deeply against GameLift and later wants to run on Azure or Google Cloud for a specific region faces a genuine re-engineering project, not a configuration change.
Google Cloud Agones: Strengths and Trade-offs
Agones’s biggest structural advantage is that it has no license fee and no vendor relationship at all, it is Kubernetes-native open source software maintained on GitHub, and it runs anywhere a Kubernetes cluster runs. That includes GKE, Amazon EKS, Azure AKS, or a studio’s own bare-metal cluster. For a studio that already has Kubernetes expertise in-house, Agones is close to a pure infrastructure decision rather than a vendor decision.
The September 2026 release, Agones 1.61.0, specifically targeted reliability issues that matter for production game fleets: a hardened SDK sidecar security context, a GameServer health-detection fix, and a resolved memory leak, according to the project’s own release notes. Combined with Kubernetes 1.34-1.36 compatibility, that signals an actively maintained project rather than an abandoned one, an important signal given that Agones has no commercial parent company backing it the way GameLift has AWS.
The cost of that flexibility is operational ownership. There is no AWS or Google support line for “why is my Agones Fleet Autoscaler not scaling fast enough,” that’s a debugging session for the studio’s own platform engineers. Teams without existing Kubernetes operational experience should budget real ramp-up time, not just compute costs, before counting on Agones to be cheaper in practice than a fully managed alternative.
Multiplay by Rocket Science: Strengths and Trade-offs
Rocket Science’s pitch is continuity. Per the company’s own announcement, it transitioned the Multiplay business “several months” before going public about it and already supports more than 60 live clients, including what it describes as a successful global, multiplatform launch. For a studio with a working Multiplay integration, Rocket Science is the only one of the three options that does not require rewriting server-fleet integration code, since the underlying platform and APIs are the same ones Unity operated before the handoff.
That continuity is also the risk. Rocket Science is a smaller, independent company rather than a division of a hyperscaler, and its pricing is not public. Studios considering this path should ask direct questions about long-term roadmap commitments, SLA terms, and what happens if Rocket Science itself is acquired or changes direction, the exact scenario that triggered this whole migration wave in the first place. Competing platforms like Edgegap have specifically built migration tooling aimed at studios weighing whether to trust the legacy Multiplay stack under new ownership versus starting fresh elsewhere.
Other Migration Options Beyond the Big Three
GameLift, Agones, and Rocket Science are the three paths most directly tied to the Unity Multiplay shutdown story, but they are not the only game server hosting options a studio should put on its shortlist. Specialized third-party providers built specifically for multiplayer game hosting have used the Unity Multiplay disruption as an opening to pitch migration support of their own. Three names come up repeatedly in that conversation: Edgegap, i3D.net, and Hathora.
| Provider | Positioning | Relevant to Unity Multiplay migration |
|---|---|---|
| Edgegap | Distributed edge game server hosting with a Unity integration plugin | Publishes a direct comparison against Rocket Science’s Multiplay, positioning itself as an alternative for studios wary of the new ownership |
| i3D.net | Dedicated game server hosting provider with its own bare-metal network | Commonly cited alongside GameLift and Agones as a third-party alternative for studios that want a specialized hosting partner rather than a hyperscaler |
| Hathora | Managed serverless game server platform aimed at indie and mid-size studios | Positioned as a lower-operational-overhead alternative for teams that want managed hosting without AWS or Google account complexity |
None of these three have published specific head-to-head pricing against GameLift, Agones, or Rocket Science as of October 2026, so they are listed here as shortlist candidates rather than fully scored contenders in this comparison. What they share is a pitch aimed squarely at the gap Unity’s exit created: studios that want dedicated game-server hosting expertise without defaulting to a general-purpose cloud platform’s container or compute primitives. A studio running a full procurement process should request quotes from at least one of these alongside AWS, Google, and Rocket Science before finalizing a decision, since specialized providers sometimes beat hyperscaler pricing specifically on dedicated-server workloads where generic cloud compute carries overhead that a purpose-built network does not.
Cost Modeling: Three Player-Count Scenarios
Published hourly rates are useful, but they don’t mean much until they’re applied to an actual fleet size. The walkthrough below applies the publicly documented rates from earlier in this comparison to three illustrative player-count scenarios. These are modeling exercises built from each platform’s own published pricing, not case studies of real studios, and they deliberately hold the underlying assumptions constant so the comparison stays apples-to-apples.
Assume a dedicated-server game where each server instance handles 20 concurrent players, running continuously (730 hours per month), using the cheapest published On-Demand compute tier for each platform.
- Small scale, 200 concurrent players (10 server instances): On GameLift’s c6g.large rate of $0.088/hour, 10 instances running 730 hours costs roughly $642/month in compute before any Spot discount. On GKE Autopilot at the low end of its published range ($0.0133/vCPU-hour for a 2-vCPU pod), the same 10-instance footprint runs closer to $194/month in raw compute, plus the $0.10/cluster-hour management fee (roughly $73/month) for a combined total near $267/month, before factoring in the engineering time to operate that cluster.
- Mid scale, 2,000 concurrent players (100 server instances): GameLift compute scales to roughly $6,424/month at the same rate before Spot savings; applying the advertised 50% low-end Spot discount brings that closer to $3,212/month. The equivalent GKE Autopilot footprint runs roughly $1,940/month in compute plus the flat cluster fee, landing near $2,013/month, again excluding operational labor.
- Large scale, 10,000 concurrent players (500 server instances): GameLift compute reaches roughly $32,120/month before Spot, or as low as $4,818/month if the full 85% advertised Spot discount applies and the game’s architecture tolerates interruption. The GKE Autopilot equivalent lands near $9,700/month in compute plus the cluster fee, a reminder that Autopilot’s advertised low-end regional rate does not automatically apply everywhere, and that preemptible Kubernetes nodes carry their own interruption risk analogous to GameLift Spot.
Rocket Science cannot be modeled this way because its pricing is not public, which is itself a data point: a studio cannot run this kind of back-of-envelope comparison against Rocket Science without first getting on a call with their sales team. That single fact should factor into how quickly a studio can make a confident decision under the migration deadline pressure this whole situation created.
Migration Guide: Moving Off Unity Multiplay
Whichever destination a studio picks, the migration mechanics follow a similar sequence. Here is the practical path.
- Audit current Multiplay usage. Document every fleet, build, region, and SDK call your game makes against the legacy Multiplay API before touching anything. This becomes your migration checklist.
- Decide your operational appetite first, not your budget first. If your team has no Kubernetes experience and no plan to hire for it, cross Agones off the list regardless of its compute pricing, the hidden cost is engineering time, not cloud spend.
- Request a Rocket Science quote in parallel with testing GameLift. Because Rocket Science pricing is not public, you cannot make an informed cost comparison without an actual quote. Run this in parallel with a GameLift proof-of-concept rather than sequentially, to avoid losing weeks waiting on a sales cycle.
- Build a GameLift or Agones proof-of-concept with a non-production build. Stand up a single fleet, run a load test with simulated concurrent sessions, and measure actual scale-up latency in your specific region before committing.
- Re-architect for session recovery if you plan to use Spot or preemptible capacity. Both AWS Spot and Kubernetes preemptible nodes can reclaim capacity mid-session. Your matchmaking and reconnect logic needs to handle that gracefully before you flip on the cost savings.
- Test free-bandwidth eligibility on GameLift. Confirm your target instance types are Generation-6 or newer to actually receive AWS’s free in/out bandwidth; older instance generations do not qualify and will add an unexpected line item to your bill.
- Run a parallel soft launch before cutting over fully. Route a small percentage of live traffic to the new platform while the legacy Multiplay fleet (if still available to you under the migration exception) continues serving the majority, then ramp gradually.
- Decommission the legacy fleet only after a full billing cycle of stable metrics. Watch error rates, p95 session-start latency, and cost-per-player for at least one full month on the new platform before fully retiring the old integration.
Pros and Cons Summary
AWS GameLift Servers
- Pro: Fully managed, no cluster or control-plane operations required
- Pro: Free bandwidth on Gen-6+ instances removes a volatile cost line item
- Pro: GameLift Anywhere supports hybrid and bring-your-own-hardware deployments
- Pro: Spot instances offer a documented 50-85% discount versus On-Demand
- Con: AWS-specific SDK and session APIs create real vendor lock-in
- Con: Published On-Demand instance pricing starts at $0.088/hour even before storage, load balancing, and data transfer outside the free-bandwidth tiers
Google Cloud Agones
- Pro: No license fee; fully open source and actively maintained (v1.61.0, September 2026)
- Pro: Portable across any Kubernetes cluster, including multi-cloud and on-prem
- Pro: Lowest theoretical compute floor via GKE Autopilot’s regional vCPU-hour pricing
- Pro: No single-vendor shutdown risk, the project cannot be deprecated by one company’s business decision the way Unity Multiplay was
- Con: No managed support line; your team owns every operational incident
- Con: GKE cluster management fee ($0.10/cluster-hour) and node costs stack on top of the free software
- Con: Requires existing or newly hired Kubernetes expertise to run safely in production
Multiplay by Rocket Science
- Pro: Direct continuity with legacy Unity Multiplay integrations, no SDK rewrite required
- Pro: Claims 60+ live clients and at least one global multiplatform launch as of mid-2026
- Pro: Fully managed, similar operational simplicity to GameLift
- Con: Pricing is not public; requires a direct sales quote to evaluate cost
- Con: Smaller independent operator carries more long-term roadmap and business-continuity risk than a hyperscaler product
- Con: Less third-party documentation and community support than AWS or Agones
Use-Case Recommendations
- Choose AWS GameLift Servers if: your team wants a fully managed product, is already on AWS for other services, and needs the free-bandwidth and Graviton cost advantages for a bursty, high-concurrency game.
- Choose Google Cloud Agones if: your studio already runs Kubernetes for other backend services, values long-term platform independence over short-term convenience, and has the engineering headcount to operate a cluster in production.
- Choose Multiplay by Rocket Science if: you have a live Unity Multiplay integration today, limited bandwidth to re-architect server-fleet code before your next content release, and want continuity above all else.
- Choose a hybrid GameLift Anywhere approach if: you need AWS’s fleet-management tooling but want to keep some servers on existing hardware or in regions where AWS pricing is less competitive.
- Choose to delay migration and request quotes from all three if: your legacy Multiplay fleet still qualifies for the migration-intent exception and you have time before that exception expires, use the window to properly benchmark rather than migrating under deadline pressure.
The Verdict: Which Path Makes Sense in 2026
There is no single winner here, and any comparison that claims otherwise is oversimplifying a decision that genuinely depends on team composition. But the data points to a reasonably clear default logic. If cost predictability and minimal operational overhead matter most, AWS GameLift Servers is the safer default, its free-bandwidth policy on Gen-6+ instances and documented Spot discounts of 50-85% give studios real, published numbers to model against, something neither Agones’ total operating cost nor Rocket Science’s private pricing can offer without further legwork.
If a studio already has Kubernetes expertise in-house and wants to never face a repeat of the Unity Multiplay shutdown scenario, where a single company’s business decision forces an urgent platform migration, Agones is the structurally safest long-term bet, because open source software under an active release cadence (v1.61.0 as of September 2026) cannot be unilaterally deprecated by a vendor’s board decision. The trade-off is accepting full operational ownership, including the GKE cluster management fee and node costs that sit on top of the free software.
Multiplay by Rocket Science is the right call almost exclusively for studios with an existing, working Multiplay integration who want to avoid an SDK rewrite under deadline pressure. Its claimed 60+ live clients and a global multiplatform launch suggest it is operationally real, not a shell migration target, but the lack of public pricing means no studio should treat it as a drop-in replacement without getting a signed quote first. Given that this entire situation started because one platform’s ownership changed, any studio choosing Rocket Science should weigh that same risk before committing fully.
Frequently Asked Questions
When exactly did Unity Multiplay Game Server Hosting shut down?
Unity deprecated Multiplay Game Server Hosting on April 1, 2026, according to its own status notice. Customers had until March 31, 2026 to declare their intent to migrate to Multiplay by Rocket Science in order to qualify for a continued-access exception while their migration completed.
Is Multiplay by Rocket Science the same company as Unity?
No. Rocket Science Group is an independent company, founded by former members of the original Multiplay engineering team, that licensed the Multiplay software from Unity and now operates it as a separate commercial platform.
Is AWS GameLift Servers cheaper than running my own Agones cluster?
On raw published rates, GKE Autopilot’s lowest regional vCPU-hour pricing (as low as $0.0133) can undercut GameLift’s cheapest listed instance type ($0.088/hour for a c6g.large). But that comparison ignores the GKE cluster management fee of $0.10 per cluster-hour and the engineering time required to build, secure, and operate your own Agones deployment. For most teams without existing Kubernetes operations, GameLift’s fully managed model is cheaper in total cost once engineering time is counted.
Can I use AWS Fargate instead of EC2 instances with GameLift or Agones?
Agones can run on Fargate-backed Kubernetes pods, and AWS Fargate separately bills per vCPU and GB-hour for containerized workloads on ECS or EKS. However, GameLift Servers itself is priced and billed as EC2 instance-hours, not as a Fargate product. Studios considering a containerized approach outside of GameLift should evaluate Fargate’s per-second billing model, which starts counting from the moment a container image begins downloading.
What happens to my existing game server fleets if I don’t migrate in time?
Per Unity’s status notice, customers who did not clearly indicate migration intent to Rocket Science before the March 31, 2026 cutoff lost the ability to scale new game servers or make new allocations on the legacy Multiplay service as of April 1, 2026. Studios in that position should treat any remaining legacy capacity as a wind-down resource, not a stable long-term platform.
Does Google charge anything for using Agones itself?
No. Agones is free, open source software with no license fee. The costs come entirely from the underlying Kubernetes infrastructure, in Google’s case the GKE cluster management fee plus standard Compute Engine node pricing, or equivalent costs if you run Agones on a different Kubernetes distribution.
Which platform is best for a small indie studio with one infrastructure engineer?
A fully managed option, AWS GameLift Servers or Multiplay by Rocket Science, is almost always the better fit for a small team. Agones demands ongoing Kubernetes operational expertise that a one-person infrastructure team will struggle to sustain alongside everything else a small studio needs from that engineer.
Are Spot instances and preemptible nodes safe to use for live multiplayer matches?
They are safe only if the game’s matchmaking and session logic is built to handle sudden capacity reclamation. AWS advertises GameLift Spot savings of roughly 50-85% versus On-Demand pricing, and Kubernetes preemptible nodes offer similar savings for Agones deployments, but both can be reclaimed mid-session. Studios running stateful, non-recoverable matches should stick to On-Demand or standard node pools, or build robust reconnect logic before adopting discounted capacity.
