Self-Hosted AI Deployment in Stockholm, Sweden

Gaming / Interactive Entertainment / Self-Hosted AI / Stockholm

A Stockholm-based gaming group (SEK 8.4B revenue — approximately EUR 740M, operating 6 game studios across Stockholm, Malmoe, and Uppsala, developing and operating 14 live-service games across PC, console, and mobile, 2,200 employees, 82 million monthly active players globally) needed to deploy AI capabilities to transform game development, live operations, and player experience. The group's games generated billions in lifetime revenue — the AI models that governed game economies, player engagement, and content generation represented the most valuable IP in the company. Core challenges: (1) Game AI as existential IP — the group's most valuable games were live-service titles (games-as-a-service model) where ongoing AI-driven operations determined revenue. Monetisation AI: models predicting which items to offer to which players at what prices — optimising revenue while maintaining player satisfaction. These models, trained on billions of player transactions, encoded the company's understanding of player spending behaviour. If a competitor accessed these models, they could replicate the monetisation strategies that generated SEK 3.2B annually. Engagement AI: models predicting player churn and recommending retention interventions (special events, content releases, difficulty adjustments). These models understood player motivation at a granular level — which players would return with a difficulty reduction versus a social feature versus a content preview. Game economy AI: models managing virtual economies in games with millions of active players — controlling currency inflation, item rarity, resource generation rates, and marketplace pricing. Economic model failures caused player outrage and revenue collapse (historical examples across the industry proving this repeatedly). Anti-cheat AI: models detecting cheaters through behavioural analysis — identifying aiming assistance, speed manipulation, economy exploitation, and account farming from player behaviour patterns. Self-hosted deployment was essential because: if cheat developers analysed the AI models (possible through API observation), they could develop evasion techniques, and if the AI was hosted externally, the external provider's employees could theoretically observe model behaviour. (2) Content generation AI — the next frontier. The group was investing in AI-generated content to: create game levels (procedural generation of dungeons, quests, landscapes, and encounters), generate NPC dialogue (AI-powered NPCs that could converse naturally with players rather than repeating scripted lines), create game assets (generating textures, 3D models, animations, and sound effects to supplement human-created content), and personalise experiences (generating different game experiences for different player segments — varying difficulty, narrative paths, and challenges based on player behaviour). Content generation AI was trained on the group's proprietary game assets — character designs, world art, narrative archives, and sound libraries representing hundreds of millions in creative investment. Sending this data to external AI providers risked: IP exposure (the art style, narrative elements, and design philosophy that differentiated the group's games), copyright concerns (the group needed to own all generated content — external AI processing creating ownership uncertainty), and quality control (game content generation requiring precise aesthetic control that only self-hosted models fine-tuned on the group's specific art style could provide). (3) Scale — 82 million monthly active players. The group processed: 4.2 billion in-game events daily (player actions, transactions, progression events, social interactions), 180 million AI inference requests daily (real-time matchmaking, content recommendations, monetisation decisions, anti-cheat checks), and peak loads of 850,000 concurrent players (requiring sub-100ms inference for real-time game experience decisions). This scale at cloud AI API pricing was prohibitive — estimated SEK 42M monthly (SEK 504M annually) versus self-hosted costs of approximately SEK 8M monthly. (4) Player data privacy — the group's 82 million players included significant numbers of children (some games rated PEGI 7 or PEGI 12). GDPR Article 8 (children's consent) and Sweden's particularly strict approach to children's data protection (IMY having investigated multiple companies for children's data processing) required: minimised data collection, purpose-limited processing, and enhanced security. AI processing player behaviour data — including children's — needed to demonstrate IMY compliance. Self-hosted deployment with no third-party data sharing provided the strongest privacy position. (5) Real-time requirements — many AI decisions needed sub-100ms latency. Matchmaking: assigning players to game sessions based on skill, geography, play style, and social preferences — decisions made in real time as players entered matchmaking queues. Anti-cheat: real-time behavioural analysis during gameplay — detecting aim assistance, speed manipulation, and economy exploitation within the game session. Dynamic difficulty: adjusting game challenge in real time based on player performance — preventing frustration (too hard) or boredom (too easy). Monetisation: in-game store recommendations presented during gameplay — contextually relevant offers appearing at optimal moments. Cloud API round-trip latency (50-200ms) was incompatible with these real-time requirements. Self-hosted inference in the group's game server infrastructure provided the sub-100ms latency needed.

We deployed a self-hosted AI platform over 10 weeks — integrated with the group's game server infrastructure across Stockholm data centres. (1) Gaming-optimised GPU infrastructure. Hardware: 48 NVIDIA H100 80GB GPUs across 6 server nodes, deployed across the group's two Stockholm-area game server facilities (facilities already hosting game servers for European players — co-located AI reducing network latency to zero for AI-assisted gameplay decisions). Game operations cluster: 32 GPUs (4 nodes) handling real-time inference — matchmaking, anti-cheat, dynamic difficulty, and monetisation decisions. Sub-100ms latency requirement met through co-location with game servers. Development and training cluster: 16 GPUs (2 nodes) for model training, fine-tuning, and content generation. Separated from production to prevent training workloads from affecting real-time game experience. Network: InfiniBand between GPU nodes. Connected to game server clusters via dedicated 100Gbps Ethernet. Game-to-AI inference path: under 5ms network latency (co-located infrastructure). Storage: 60TB NVMe for model weights and active game data. Connected to the group's game data lake (2.8PB — player behaviour data, game telemetry, asset libraries). Auto-scaling: inference workload scaling with player population — GPU allocation automatically adjusting between games based on concurrent player counts (Kubernetes-based orchestration with NVIDIA GPU Operator). (2) Game economy and monetisation AI. Economy management: Llama 3 70B fine-tuned on the group's game economy data — learning the relationship between economy parameters (currency generation rates, item drop rates, marketplace pricing, crafting costs) and player behaviour (engagement, spending, progression, satisfaction). The model monitoring live game economies in real time: detecting inflation (currency accumulating faster than intended — indicating exploit or design flaw), identifying economy imbalances (specific items becoming too common or too rare — disrupting game balance), and recommending parameter adjustments (economy tuning suggestions for game designers, presented through an internal dashboard). Economy AI reducing emergency economy interventions from 14 per quarter to 3 (developer team intervening only for AI-flagged critical issues rather than reacting to player complaints). Monetisation optimisation: models recommending: which items to feature in the in-game store for each player segment (based on play style, progression stage, and historical purchase patterns), optimal pricing (dynamic pricing within ranges approved by the game design team — not gouging but matching willingness-to-pay), and timing (when during a play session to present offers — after positive experiences, not during frustrating moments). Monetisation AI increasing per-player revenue 18 percent while maintaining player satisfaction scores (measured through in-game surveys and engagement metrics) — demonstrating that well-targeted offers improved rather than degraded player experience. (3) Anti-cheat AI: protecting competitive integrity. Behavioural analysis: Mistral 7B fine-tuned on the group's cheat detection dataset (84,000 confirmed cheat cases with behavioural profiles) — detecting: aim assistance (statistical analysis of aiming patterns — accuracy, snap-to-target speed, reaction time distributions that differed from human capabilities), speed and movement manipulation (movement patterns exceeding normal parameters — wall clipping, speed increases, flight), economy exploitation (resource generation patterns indicating automated farming, duplication exploits, or marketplace manipulation), and account sharing/selling (behavioural pattern changes indicating different humans using the same account). Real-time processing: analysing player behaviour during gameplay at sub-50ms latency. The model scoring every active player continuously, flagging suspicious behaviour for investigation. Key advantage of self-hosted: the anti-cheat model was completely opaque to external analysis. Cheat developers could not observe the model's behaviour through API probing (as they could with cloud-hosted models) because the model ran entirely within the game server infrastructure with no external interface. Results: cheat detection rate from 72 percent to 94 percent (22 percentage point improvement). False positive rate maintained at 0.3 percent (falsely banning legitimate players being extremely damaging to reputation). Time to detect: from average 48 hours (manual review of reports) to average 12 minutes (real-time behavioural detection). Competitive game player satisfaction increased 34 percent post-implementation (measured through player surveys about match quality). (4) Content generation AI: accelerating game development. Level generation: fine-tuned models generating game levels, dungeons, and environments based on the group's design specifications and art style. Game designers providing high-level briefs ("a medium-difficulty dungeon for level 30-35 players with a frost theme and a boss encounter") and the AI generating: spatial layouts (room configurations, corridors, environmental features), encounter placement (enemy types, positions, patrol routes, difficulty scaling), reward distribution (loot tables, resource nodes, secret areas), and visual styling (texture selection, lighting parameters, environmental effects matching the game's art direction). Human designers reviewed and refined AI-generated content — the AI producing "first drafts" that designers iterated on rather than building from scratch. Content generation productivity: level designers producing 3.4x more content per sprint (measured in completed, quality-approved game content). NPC dialogue: Llama 3 70B fine-tuned on the group's narrative archives (4.2 million lines of dialogue across 14 games) — generating NPC dialogue that matched each game's narrative tone and world-building. NPCs conducting dynamic conversations with players rather than repeating scripted lines — responding to player questions, commenting on game events, and revealing lore organically. Player feedback: 76 percent of surveyed players rating AI-generated NPC dialogue as "good" or "excellent" quality — compared to 82 percent for human-written dialogue (close to parity). Asset generation: AI generating supplementary game assets — textures, simple 3D props, and sound effects that supplemented human-created core assets. Asset generation reducing art team workload on repetitive assets by 28 percent, allowing artists to focus on hero assets and creative direction. (5) Player privacy and children's data protection. Data minimisation: AI models designed to process behavioural patterns rather than personal identifiers — player behaviour represented as feature vectors rather than linked to personal data. This approach satisfying GDPR data minimisation principles while enabling effective AI. Children's data: for games with significant underage player populations, enhanced protections: no monetisation AI targeting for accounts identified as under-16 (parental controls and age verification), content moderation AI more conservative for children's game modes, and behavioural data processing limited to game experience optimisation (not monetisation or profiling). IMY compliance: self-hosted deployment with no third-party data sharing providing the strongest privacy position for IMY scrutiny. Complete data processing records maintained for every AI application, demonstrating purpose limitation and proportionality.

Results over 12 months post-deployment. Game operations: monetisation optimisation: 18 percent increase in per-player revenue across live-service titles (SEK 576M additional annual revenue on SEK 3.2B baseline). Player satisfaction maintained or improved — validating that better-targeted offers improved experience. Game economy management: emergency interventions from 14 to 3 per quarter (78 percent reduction). Economy stability contributing to 12 percent improvement in 30-day player retention. Anti-cheat: detection rate from 72 percent to 94 percent. False ban rate maintained at 0.3 percent. Competitive game player satisfaction up 34 percent. Estimated player retention impact: 8 percent fewer players leaving due to cheating frustration (representing approximately SEK 180M in retained annual revenue). Content generation: level design productivity: 3.4x content output per sprint. Content pipeline acceleration enabling the group to maintain live-service content cadence with 15 percent smaller content team (redeployment to new projects rather than headcount reduction). NPC dialogue quality: 76 percent "good or excellent" rating (approaching human-written 82 percent). Asset generation: 28 percent reduction in repetitive asset workload. Overall content production cost reduction: SEK 42M annually (through productivity improvement, not quality reduction). Player metrics: monthly active players: from 82 million to 94 million (15 percent growth — partially attributed to improved game quality through AI content and reduced cheating). Player retention (30-day): from 42 percent to 54 percent (12 percentage point improvement). Average revenue per user (ARPU): from SEK 39 to SEK 46 monthly (18 percent increase). Performance: H100 cluster handling peak loads of 180 million daily inference requests. Real-time inference latency: median 28ms for matchmaking, 42ms for anti-cheat, 65ms for monetisation (all within sub-100ms requirement). System uptime: 99.97 percent (gaming infrastructure requiring near-perfect uptime — downtime during peak hours costing approximately SEK 800K per hour in lost revenue). Cost: infrastructure investment: SEK 48M (approximately EUR 4.2M — 48 H100 GPUs, servers, networking, storage). Annual operating cost: SEK 9.6M (EUR 845K). Annual value: monetisation improvement SEK 576M, anti-cheat retention SEK 180M, content production savings SEK 42M. Total annual value: SEK 798M (approximately EUR 70M). ROI: 13.8x in first year. The group's CEO noted that the self-hosted AI platform "transformed our competitive position — we can now iterate on game AI at a pace our competitors cannot match because they are constrained by cloud API limitations and IP concerns."