Q: What is the scale and timeline of HoYoverse’s AI plan? A: HoYoverse plans to spend up to $14.6 billion on AI over the next three years to build internal tools, GPU clusters, and training systems. The HoYoverse in-house AI investment was outlined in a private technology and recruitment session on May 15 and focuses on creating a full internal AI stack rather than relying only on external models. Q: What core components will HoYoverse build as part of its in-house AI stack? A: It covers GPU clusters for compute, training systems including data pipelines, labeling, fine-tuning and evaluation loops, and application architecture such as model gateways, inference services, prompts, memory, and safety filters. The HoYoverse in-house AI investment also includes observability for cost, latency and quality, plus security and privacy guardrails to protect IP and user data. Q: How will this investment affect NPCs and live-service features in games? A: The plan aims to enable smarter NPC behavior with dynamic dialogue, schedules, and reactions, and to automate content generation and live-service tuning like personalized events and offers. The HoYoverse in-house AI investment is expected to be applied in upcoming titles such as Petit Planet to power AI-driven NPCs and scaled live ops features. Q: What build-or-buy lessons should developers take from HoYoverse’s approach? A: The article advises building internal systems when features are strategic—examples include NPC brains and proprietary live ops—and buying when features are commodity, like OCR, basic TTS, and standard translation. The HoYoverse in-house AI investment underscores tracking total cost of ownership for compute, staff, and data, owning data pipelines, and designing for portability to avoid lock-in. Q: What immediate practical steps can studios take now in response to this trend? A: Studios should inventory and clean their text, audio, art, and telemetry, fix rights and storage, and label clean training and fine-tune sets with version control. The HoYoverse in-house AI investment guidance also recommends prototyping small tools that remove pain points, measuring quality and cost per call, and setting latency budgets and a model gateway before scaling. Q: What are the main risks and how can studios de-risk building in-house AI? A: Main risks include high compute burn and cost per feature, model drift that requires regular re-testing and model version locks, legal exposure over training data rights, and potential loss of player trust for sensitive interactions. The HoYoverse in-house AI investment materials suggest de-risking by capping training runs, running human-in-the-loop reviews for story and live events, and providing fallback behavior when models fail. Q: Which signals should developers watch to monitor HoYoverse’s AI progress? A: Developers should monitor hiring patterns for AI engineers, data ops and applied researchers, GPU procurement and infra scale, and the emergence of internal tools in games. The HoYoverse in-house AI investment also points to Petit Planet as a testbed for NPC memory and live events and to possible open-source releases or model-serving frameworks as signals. Q: How should a studio decide whether to build internal AI tools or rely on external vendors? A: Studios should evaluate strategic importance, whether vendor models meet feature and latency needs, and the total cost of ownership including compute, staff, data, and upkeep before choosing to build. The HoYoverse in-house AI investment example recommends renting burst compute when possible, caching results to save cost, and owning data pipelines where IP or unique design requires it.