Cloud gaming finally has a business model that works, and it's not what Stadia tried
Google shut down Stadia in January 2023 after spending an estimated $1 billion-plus trying to convince people to replace their consoles with a streaming service. Three years later, Nvidia's GeForce Now and Microsoft's Xbox Cloud Gaming are both profitable-adjacent, growing, and increasingly treated as a default feature rather than a novelty. The difference isn't better video codecs or lower latency, though both helped. It's that the surviving services stopped trying to solve the problem Stadia tried to solve.
Stadia asked a gamer to abandon the thing they already owned — a library of games on Steam, a PlayStation, an Xbox — and rebuild it from scratch on Google's platform, paying full price for titles a second time. That's not a streaming pitch. That's a migration with extra latency risk, and gamers correctly treated it as a worse deal than just buying a console.
The model that actually works: stream what you already own
GeForce Now's entire pitch is the opposite: bring your own games. If you own a title on Steam, Epic, Ubisoft Connect, or EA's launcher, GeForce Now streams it to you on a phone, a Chromebook, a smart TV, or a gaming PC you don't have to upgrade. Nvidia never had to negotiate exclusive streaming rights for most of its 2,000+ supported titles, because it isn't distributing the game — it's renting you the GPU that runs a copy you already licensed.
This single structural choice changes everything downstream. Nvidia doesn't carry publisher licensing costs the way Stadia did when it bought ports and timed exclusives. It doesn't need to convince EA or Ubisoft to trust a new distribution channel with cannibalization risk. And it doesn't ask the player to make a binary bet that cloud gaming will still exist in three years before they'll buy anything — they've already bought the game regardless of how they play it.
Xbox Cloud Gaming took a related but distinct path: it bundled cloud streaming into Game Pass Ultimate rather than selling it as a standalone product. That reframes cloud gaming from "a new way to buy games" to "a value-add on a subscription you were already paying for," which is a much easier sell. Microsoft doesn't need cloud gaming to be profitable in isolation; it needs it to reduce churn on Game Pass, which it does by letting subscribers play Xbox library titles on a phone during a commute without owning a console at all.
The cost side finally caught up
The other half of the story is unglamorous infrastructure economics. A GPU streaming session costs real money per hour — data center GPU time, bandwidth, encoding overhead. Early cloud gaming services ran near-dedicated capacity per user because demand was unpredictable and session scheduling was primitive. That meant paying for idle GPU time between sessions, which made the unit economics brutal regardless of subscription price.
Modern cloud gaming platforms share GPU capacity across time-sliced sessions more efficiently, use better session-length prediction to pre-provision capacity, and benefit from falling per-FLOP GPU costs even as demand for AI training has pushed up GPU prices elsewhere — cloud gaming providers lock in older-generation GPU capacity at a discount once AI workloads move to newer silicon. AV1 and improved H.265 encoding cut bandwidth costs by double-digit percentages compared to the codecs Stadia launched with, which matters enormously when you're serving millions of concurrent 1080p-4K streams.
Network conditions also improved independent of anything a gaming company did: 5G coverage expanded, fiber-to-the-home penetration grew, and round-trip latency on typical home connections dropped enough that 30-50ms added latency from cloud rendering became tolerable for most genres outside competitive shooters and fighting games — which, notably, neither GeForce Now nor Xbox Cloud Gaming markets itself heavily around.
What this means going forward
The lesson generalizes past gaming: distribution-layer products that ask users to abandon an existing library or workflow face an adoption tax that's almost impossible to overcome with subscription pricing alone, no matter how good the underlying technology is. Products that slot into an existing purchase — streaming what you already bought, running on top of a subscription you already pay for — inherit demand instead of having to manufacture it.
For publishers, the BYOG model also quietly solves a problem they didn't ask anyone to solve: it extends the addressable hardware market for every PC game they sell. A title that requires a $2,000 GPU to run natively can now reach a Chromebook or a budget phone, and the publisher collects the same unit sale either way, with Nvidia absorbing the hardware cost as a service fee instead.
The takeaway for anyone evaluating cloud infrastructure plays, gaming or otherwise: the winning product rarely tries to disintermediate the thing people already paid for. It finds the layer where it can add reach without asking for a second payment, and lets someone else's existing demand do the hard work of customer acquisition.