Cloud gaming runs Windows VMs. Should it?

Game Pass Cloud, GeForce Now Ultimate, PS Plus Premium streaming, Boosteroid, Luna — all of them run a copy of Windows 10 or Windows 11 on cloud GPU instances. The user's cloud session is functionally a Windows VM with a GPU passed through, running games as Windows binaries.

PS Plus Premium streams PS5 hardware emulation rather than Windows in some session configurations, but the broader pattern across the industry is overwhelmingly Windows-on-cloud-GPU.

Historical accident, mostly. When cloud gaming services were designed in 2018-2020, Windows was the dominant gaming OS. Running Windows in the cloud was the lowest-friction way to support the existing Windows-binary game catalogue.

The architecture choice locked in a set of trade-offs: Windows licensing costs, Windows-specific performance characteristics, Windows update cadence (which causes occasional cloud session disruption when major updates roll out), and Windows-specific security models.

Alternatives existed in 2020 — pure-Linux gaming environments running through Proton, custom OS implementations on cloud GPUs. None scaled because the AAA catalogue is Windows-binary.

Microsoft Windows licensing. Each cloud GPU instance is paying a Windows datacenter SKU license to Microsoft. For Game Pass Cloud this is intra-Microsoft accounting. For everyone else, it's revenue going to a direct competitor.

Performance overhead. Windows has substantial OS overhead even with modern optimizations. Headless Linux on the same GPU would deliver 5-15% better gaming performance for the same hardware. Cloud services eat this overhead and pass the cost to users.

Update fragility. Windows monthly updates occasionally break specific games or cause session disruption. Cloud services have to manage Windows update timing carefully and sometimes delay critical patches.

Game catalogue compatibility. Moving to Linux+Proton or a custom OS would require validating every game in the catalogue. The effort is substantial.

Inertia. Cloud gaming services are built on Windows, the engineers are trained on Windows, the tooling is Windows-aware. Migration costs are large and the incremental benefit (better performance, lower licensing) is not viscerally large enough to justify it.

Microsoft strategic posture. Microsoft wants Windows to remain the gaming OS. Windows datacenter licensing terms favor cloud gaming services that stay on Windows. Moving away from Windows would forfeit that pricing benefit.

NVIDIA strategic interest in Linux gaming. NVIDIA has invested heavily in Linux gaming through Proton, the Steam Deck partnership, and various other moves. If NVIDIA decided to run GeForce Now on a Linux-based stack to reduce licensing costs and improve performance, the engineering investment is plausible.

Sony, conversely, has incentive to run more of PS Plus Premium streaming on actual PS5 hardware (which runs Sony's BSD-based OS) rather than Windows VMs. Their architecture is already partially BSD-based.

Microsoft has zero incentive to move Game Pass Cloud off Windows. The licensing economics flow within their own company.

Headless Linux running Proton-translated Windows binaries. Roughly 95% catalogue compatibility on launch, with a list of exceptions that gets actively addressed by the service operator.

Or: a custom thin-OS layer specifically designed for game execution, with a Wine-equivalent compatibility layer underneath. More performant than Windows + Hyper-V, but more engineering investment.

Or: native Linux versions of titles where available, with cloud-streaming-specific patches. Limited catalogue, high performance, lowest licensing cost.

None of these are realistic short-term ships. They're plausible 2028-2030 architecture moves.

Through 2027: industry stays on Windows. The status quo wins.

By 2028: NVIDIA experiments with Linux-based GeForce Now tier as a cost-reduction play. Initial deployment is limited; expansion depends on user-quality outcomes.

By 2030: cloud gaming industry is mixed Windows/Linux. Premium tiers stay on Windows for catalogue completeness; cost-optimized tiers move to Linux+Proton for the subset of titles that work cleanly.

The question 'should cloud gaming run Windows VMs' has an honest answer: probably not, but moving away from it is structurally hard, so it will keep running Windows VMs for at least 3-5 more years regardless.