At the Global Games Show 2025 in Abu Dhabi, three executives with backgrounds spanning game data platforms, distributed computing infrastructure, and AI game tooling sat down to discuss what artificial intelligence is actually doing to game production — not in theory, but in practice. Omar Ghanem is founder and CEO of GAM3S.GG, a platform that analyzes player behavior data to power game recommendations. Daniel Wang is co-founder and CEO of Aethir, which provides distributed GPU and AI compute infrastructure. Tung Nguyen-Khac is co-founder of VIZTA Digital, a studio services company working with developers on AI tooling across the production pipeline. The conversation is moderated by Georgina, MC of the event.
Personalization at scale
Omar opens with what GAM3S.GG actually does with AI: pulling data from a player’s connected accounts — games played, games played with friends, money spent — and using that to generate recommendations based on demonstrated behavior rather than assumptions about what someone might want. The distinction he draws is between what a platform thinks you should buy and what your actual patterns suggest you’d enjoy. At human-analyst scale, this kind of individualized profiling would take an impractical amount of time per user. AI makes it continuous and automatic across millions of players simultaneously.
The underlying shift here is significant for how game discovery works. Historically, game storefronts have surfaced content based on sales volume, editorial picks, or broad demographic targeting. Behavior-based recommendation at the individual level is a different model — one that theoretically surfaces niche titles that match a player’s specific tastes rather than just surfacing whatever is most popular at that moment. Whether it works as well in practice as it does in theory depends on the quality and diversity of the underlying data, and on whether the recommendation system is genuinely player-serving or quietly weighted toward commercial priorities. Omar’s framing — “not on what we want to sell to you” — is the right aspiration. Whether any platform fully achieves it is a harder question.
The compute problem behind immersive worlds
Daniel’s contribution focuses on infrastructure rather than content, and the problem he describes is one that most players never think about: the rendering and compute demands of increasingly complex open worlds are growing faster than centralized cloud infrastructure can cheaply serve them, particularly in regions far from major data center hubs.
The solution Aethir is working on is distributed decentralized compute — essentially drawing on GPU resources from around the world rather than routing everything through a small number of large data centers. The practical benefit for players in emerging markets is reduced latency. A player in Abu Dhabi routing game interactions through a data center in Amsterdam experiences a delay — potentially 150 milliseconds or more — that a player near a local data center does not. In fast-paced games, that delay is the difference between a playable experience and an unplayable one.
The Gulf region is specifically interesting for this kind of infrastructure, Daniel argues, because the combination of available land, sunlight for solar energy, and growing player populations makes localized data center development viable in a way it might not be in more densely developed regions. This is a broader trend: the Middle East and North Africa gaming market has been one of the fastest-growing in the world, and the infrastructure investment required to serve it properly is starting to follow.
AI as co-pilot in the production pipeline
Tung’s section of the discussion is the most grounded in day-to-day production reality. His framing — AI as co-pilot rather than replacement — is the consensus position in the games industry right now, and it is worth understanding what it means in practice rather than as a talking point.
The translation problem he describes between creative vision and technical production is real and persistent. A designer or product manager has a clear mental image of what a game should feel like. Communicating that to an engineering team, then iterating when the result doesn’t match the vision, is expensive in time and money. AI tools that allow vision-holders to prototype or mock up concepts directly — through prompt-based interfaces rather than requiring engineering hand-off — compress that loop significantly. The result is not that the designer replaces the engineer, but that the initial translation step becomes faster and cheaper, which means more iterations happen before committing to a direction.
Daniel adds a dimension from his time at Riot: one of the most expensive parts of game development is the cost of changing direction. When a creative lead wants to rebalance a feature or try a new mechanic, the entire team has to reprioritize. AI tooling that allows parallel work streams — generating multiple versions of a concept simultaneously rather than sequentially — changes the economics of that decision significantly. The creative lead can choose from a lineup rather than committing to a single direction before seeing it built.
Tung is also specific about which parts of the pipeline are seeing the fastest transformation right now: asset production and pre-production. QA testing, color matching, localization, and repetitive content generation are the places where AI is delivering the most immediate productivity gains, because these are tasks that are high-volume, relatively well-defined, and not primarily dependent on creative judgment.
AI NPCs and the community tension
One of the more nuanced points in the panel concerns the tension between personalized experiences and social gaming. As AI enables increasingly tailored game content — procedurally generated environments, NPCs that respond to individual player behavior, story branches driven by personal choices — the experience diverges between players. Two people playing the same game may now have substantially different experiences.
That divergence has an upside: the game feels more personal, more responsive, more like it was made for you specifically. But it has a social cost. Part of what makes gaming communal is shared experience — the same map, the same difficult boss, the same ending. When experiences fragment significantly between players, the common reference points that allow people to compare notes, share strategies, and bond over collective moments start to erode.
Daniel’s observation is that AI NPCs are partly a solution to a different version of this problem: the need for a critical mass of active players in online spaces. Not every player has a group of four to five friends available at the same time they want to play. AI companions that behave convincingly enough to fill those gaps — and that can adapt to an individual player’s style and pace — address the loneliness of playing without a group without requiring that everyone else want the same experience at the same moment.
The line between an AI NPC and a live player is already blurring in ways the panel discusses candidly. In competitive games, bot detection has become an AI-versus-AI problem: human moderators cannot reliably distinguish well-designed bots from human players, so platforms are now using AI to detect bots by identifying behavioral signatures no human produces — the absence of the eye movement and map-swiping delay a human player would have in League of Legends, for instance. The arms race between AI performance and AI detection is already underway, and neither side is winning cleanly.
Localization as an underrated use case
Omar raises localization in a way that deserves more attention than it typically receives in AI-and-gaming discussions. Making a game available in additional languages has historically required significant resources: human translators fluent in the target language and familiar with gaming terminology, extensive QA for translated text, and ongoing work as games update. The cost has meant that many games launch in a small number of major languages and either never reach others or reach them significantly later.
Large language models have changed this calculation substantially. The quality of AI translation for major world languages is now good enough for game text in most genres, which means a studio can ship in Arabic, Mandarin, Portuguese, Hindi, and a dozen other languages at launch rather than treating localization as a future roadmap item. For players in regions that have historically received games late or not at all because of localization costs, this is genuinely significant. And for developers in those regions — the emerging market studios that Tung specifically addresses — it means their own games can reach global audiences without the localization bottleneck that previously made international distribution difficult.
Ethics and the governance gap
The panel’s treatment of ethics is honest about the difficulty without offering easy answers. The IP question is the most immediate practical concern: AI image and audio generation trained on existing creative work raises questions about when that use constitutes infringement that courts and regulators are still working through. The panel’s consensus is that governance frameworks will arrive eventually but are currently lagging the technology, which is the normal pattern for emerging technology rather than a reason for inaction.
Omar’s point about job displacement is worth sitting with. The pushback from creative communities — artists, animators, voice actors — against AI in game production is not primarily about philosophical objections to the technology. It is about whether AI tools reduce the number of people employed doing creative work, or whether they change the nature of that work while keeping people employed. The panel’s optimistic view is that AI expands what creative teams can produce rather than simply reducing how many people are needed to produce it. The pessimistic view is that the math works out differently for junior and mid-level roles, where the volume tasks that AI is best at were previously the entry points into the industry. Neither view has been definitively proven by the data yet.
For studios in emerging markets
Tung’s practical advice for studios in developing regions is to start with the repetitive and automatable rather than attempting to rebuild entire pipelines around AI at once. Use AI for asset generation, QA, localization, and documentation first. Let those productivity gains compound before making larger structural changes to how creative work is organized. The studios most likely to struggle with AI adoption are those that treat it as an all-or-nothing proposition — either fully transforming how they work or ignoring it entirely.
The broader opportunity he identifies is that AI lowers the capital requirements for game development in ways that specifically benefit studios in markets that have historically been priced out of certain types of production. A studio in Southeast Asia or the Gulf region that can use AI to accelerate pre-production, reduce localization costs, and prototype more efficiently is competing on more equal footing with larger studios in more established markets than would have been possible five years ago. The tools are not free, and expertise still matters, but the gap between what a well-resourced small team can produce and what a large studio can produce is narrowing.
The full panel is available on the Global Games Show YouTube channel. The event took place in Abu Dhabi in 2025.