Gaming powerhouse MiHoYo, known for hits like Genshin Impact and Honkai Star Rail, racked up a hefty bill in its push into artificial intelligence. An internal test of collaborative AI agents consumed nearly $300,000 worth of tokens in just 13 hours.
The Expensive Overnight Experiment
The revelation came from Zheng Yinhe, head of the AI NPC and Gameplay technology team for the Honkai series. He shared the details at the 2026 Alibaba Cloud Summit on May 20. An employee had built dozens of AI agents to work together on a single project. The goal was productive collaboration. Instead, the agents spiraled into a nonstop cycle of conversation, mutual critiques, output generation, and self-review. The process ran unchecked overnight and filled with hallucinations, yielding no usable results.
Zheng called the episode a necessary learning expense. He stated, "We accept that there are costs and learning fees when exploring AI, and this also helps to improve our agent platform." In gaming circles, the story quickly spread online. Many users joked that it was the world's most expensive group chat.
MiHoYo's Massive AI Investment
The incident highlights the steep costs of experimenting with multi-agent AI systems. Token usage can escalate rapidly when agents loop without proper guardrails. For a company the size of MiHoYo, the $300,000 hit represents a small fraction of its broader AI ambitions. Yet it serves as a concrete reminder of the financial risks involved.
MiHoYo has signaled a massive commitment to AI. Co-founder Liu Wei announced plans to invest up to 100 billion yuan (roughly $14 billion) in the technology over the next three years. He described the spending as akin to "setting off a big firework," openly accepting that not every initiative will succeed. The funds aim to build in-house tools for game development, including advanced NPCs, gameplay systems, and other creative pipelines.
Lessons for the AI Industry
This latest story arrives as the industry grapples with the realities of scaling AI. Large language models and agent frameworks promise efficiency, but uncontrolled interactions can lead to runaway expenses. MiHoYo's experience offers a case study in why companies are now focusing on better monitoring, cost controls, and platform refinements for agent-based workflows.
For fans of MiHoYo games, the news also sparks curiosity about future applications. If the company can turn these expensive lessons into smarter AI tools, players could see more dynamic characters, richer storytelling, and innovative gameplay features down the line. In the meantime, the 13-hour experiment stands as a vivid example of both the excitement and the pitfalls of frontier AI development.
As MiHoYo continues its "all in AI" strategy, the industry will be watching closely. One expensive overnight chat may not derail big plans, but it underscores a key truth: innovation in this space comes with a very real price tag.
Comments