China's AI Model Race Keeps Expanding as miHoYo, Xiaomi, and Meituan Join the Fray
nashnova research
miHoYo plans to spend up to ¥100 billion on AI over three years while Xiaomi and Meituan push their own foundation models — China's AI field is fragmenting further, driven by deep mutual distrust and distillation techniques that keep lowering the barrier to entry.
Why are three "outsiders" building their own large models?
miHoYo plans to invest up to ¥100 billion (≈$14.9 billion) in AI R&D over three years. Co-founder Cai Haoyu leads the foundation-model team personally and is seeking Nvidia Blackwell chips.
Xiaomi released its flagship model MiMo-V2.6-Pro, ranking ninth on the Artificial Analysis Intelligence Index — tied with xAI's Grok 4.7 and ahead of Alibaba's and Z.ai's latest flagships.
Meituan launched LongCat 2.0 in June. CEO Wang Xing called in-house models "the only rational strategy for the AI revolution," citing long-term cost control — but the model still trails most Chinese peers.
Why hasn't the expected consolidation happened?
After the 2023 "hundred-model war," markets expected China's AI sector to consolidate. That has not materialized. This means → fragmentation is not a transitional phase; it is the market's current steady state.
ByteDance, Alibaba, and Tencent continue pouring resources into proprietary models. Z.ai and MiniMax listed in Hong Kong in January; DeepSeek, Moonshot, and StepFun are preparing IPOs.
In plain terms = no one is dropping out, and a fresh wave of entrants and near-IPO startups keeps piling in — the field is still expanding.
Why won't anyone use someone else's model?
Bernstein senior analyst Robin Zhu notes that founders widely fear losing access to an external platform would leave them "stranded" without a model of their own.
This reflects deep mutual distrust among Chinese tech companies — each believes building AI in-house is more cost-effective than acquiring a rival.
Distillation — training your own model on outputs from frontier models such as Anthropic's and OpenAI's — is widespread. It lowers the barrier to starting from scratch and helps more players stay competitive.
How does China's landscape differ from the U.S.?
The U.S. market has settled into a clear hierarchy: OpenAI and Anthropic lead decisively, with trailing competitors at a visible distance. In China, leaderboards reshuffle with every new release, and no single company has pulled meaningfully ahead.
This means → China's AI race remains in a free-for-all stage where everyone has a chance and no one has won.
Whether capital-market pressure after the leading companies' IPOs can break this pattern is the next key question to watch.
市场有风险，内容仅供研究参考，不构成投资建议。