🧠 Food for thought

Implications, context, and why it matters.

MAU growth hides monetization gaps for China AI leaders

• China has 729 million AI app users 1. QuestMobile leaves out paying user counts, average revenue per user (ARPU) and revenue from Doubao, DeepSeek, plus Baidu’s AI search.
• DeepSeek spent $5.5 million on training while GPT‑4 topped $100 million 2. That cost gap can enable profitable unit economics (profitability per user or transaction), but without ARPU or free‑to‑paid conversion rates investors cannot judge if big audiences pay.
• Education apps such as K-12 tutoring or homework-help from Kuaidu plus Zuoyebang lead in in‑app revenue growth 3.
• About 60% of AI‑native apps shrank 1. Usage growth does not ensure retention or paid demand, so revenue metrics sort businesses from vanity stats.

Mid tier apps race to add plug and play AI as standalone apps consolidate

• In‑app AI features drew 706 million users, while AI‑native apps reached 287 million 1. Integrated features win mass adoption.
• Toolmakers plus investors can target China’s AI software development kit (SDK) market, plus the on‑device inference vendor pool. Apps want plug and play large language model (LLM) plus voice features without starting from scratch. On‑device inference means running models locally on phones or PCs rather than the cloud.
• AI agent development platforms (software that lets teams build autonomous task assistants) offer low‑code or no‑code tools 4. Developers had built over 100,000 agents by December 2024.
• Hardware makers plan for 70% AI PC penetration and 54% AI smartphone penetration by 2028 4. AI PCs plus AI smartphones have built‑in accelerators that run models locally.