Boosteroid has published three scenarios exploring how demand for new GPU infrastructure could change through 2041. The cloud gaming provider is looking beyond games to examine how AI, robotics, scientific research and entertainment could shape the demand for server-side computing power.
In its most compute-intensive scenario, graphics, gaming and media account for 25% of new GPU and accelerator infrastructure demand by 2041, up from an illustrative 13% in 2027. The other two scenarios put that share at 16% and 12%, depending on how widely AI agents and autonomous machines are adopted.
Published October 15, the analysis uses illustrative estimates rather than firm forecasts. It also explores how future games with AI characters, generated dialogue and constantly changing worlds could require more processing power, even if gaming hours stay the same.
Three Paths for New GPU Infrastructure Demand
Boosteroid’s digital-work scenario assumes businesses adopt AI agents faster than companies can deploy new robots and autonomous equipment. That would put general AI inference and agents at 46% of new GPU and accelerator demand by 2041.
The physical-AI scenario assumes faster growth in robots, autonomous transport and industrial automation. In this version of 2041, physical AI accounts for 40% of new infrastructure demand. Some processing would still happen on the machines themselves, so the amount performed in data centres remains an important variable.
In the compute-rich scenario, both automation and entertainment require more computing power. General AI inference and agents account for 30% of new demand, with physical AI at 20%. Graphics, gaming and media take a larger share here because an hour of entertainment could require more computing than it does today.
All three outcomes concern demand for new GPUs and accelerators. They aren’t estimates of how the world’s existing GPU capacity will be divided in 2041.
Model Training Takes a Smaller Share
General AI model training and development accounts for 28% of new infrastructure demand in 2027 in every scenario. By 2041, its share ranges from 9% to 12%. That doesn’t necessarily mean less computing will be used to train AI. Its absolute requirements could continue growing even as other workloads grow faster.
One reason is the expanding use of models after training. AI agents use computing power for software tasks. Autonomous machines need it for activities such as navigation and planning. The scenarios divide that demand by its end use, separating general model development from the systems that use accelerated computing.
That distinction applies to entertainment as well. AI characters and generated content could increase the processing required during gameplay. Boosteroid places graphics, gaming and media in one category, even when some of that activity involves AI.
Future Games Could Need More Computing Power Per Hour
Cloud gaming relies on remote servers to process gameplay and stream video to another device. Future game worlds could add more work to those servers through AI characters and continuous simulation.
A game world could generate character dialogue during conversations or continuously update its environment. Alongside rendering the graphics, servers might also be handling new content and simulations as events unfold.
Studios could also shift some production work to remote computing, including asset generation and simulation.
There is an important limit to the numbers. Boosteroid’s 25% scenario covers graphics, gaming and media together, rather than cloud gaming on its own. The company doesn’t provide a separate cloud gaming share or estimate how many servers future games might require.
The scenarios suggest cloud gaming’s future infrastructure needs could depend as much on computing per hour as on how many hours people play.
As always, remember to follow us on our social media platforms (e.g., Threads, X (Twitter), Bluesky, YouTube, and Facebook) to stay up-to-date with the latest news. This website contains affiliate links. We may receive a commission when you click on these links and make a purchase, at no extra cost to you. We are an independent site, and the opinions expressed here are our own.
