HoYoverse Invests $14.6 Billion in Proprietary AI Infrastructure

HoYoverse is committing $14.6 billion to build its own generative artificial intelligence infrastructure over the next three years. The company plans to bypass third-party large language model licenses in favor of vertical integration. This strategy involves constructing proprietary GPU clusters and developing custom model-training frameworks from scratch.

Studio builds custom GPU clusters and NLP models for dynamic NPC interactions

The investment targets HoYoverse’s internal game development pipeline rather than external software services. The studio aims to control every layer of the AI stack, including hardware selection and neural network architecture. This approach marks a shift away from the industry standard of licensing pre-trained models.

The first major application for this infrastructure is Petit Planet, an upcoming life simulation title. HoYoverse intends to use its proprietary natural language processing models to power dynamic conversations with non-player characters. The system will allow these interactions to change based on player actions rather than relying on static dialogue trees.

Liu Wei stated that the studio treats potential capital losses as an elaborate fireworks display if the pipeline fails to meet its objectives. This comment highlights the high-risk nature of building proprietary AI tools instead of adopting safer, licensed alternatives.