Remember when China's AI scene was a chaotic free-for-all with every tech giant and startup launching their own large language model? The so-called "hundred model war" — where Baidu, Alibaba, Tencent, ByteDance, and dozens of smaller players all raced to build the biggest, most benchmark-topping AI? That phase is ending. And according to JPMorgan, that's actually good news.
Alex Yao, head of China equity research at JPMorgan, says the landscape is rapidly consolidating from a fragmented "hundred-model" battle down to a smaller group of globally competitive players. But here's the twist: the winners won't be decided by who has the smartest model. It'll be who can turn AI into actual money.
The Einstein Problem
Yao drops a quote that should make every AI researcher pause: "You don't need a model with the intellect of Einstein. Once a model reaches the capability of a strong master's-level graduate, it can start doing real work."
In other words, China's AI companies have stopped chasing the absolute bleeding edge of model performance and started focusing on what actually matters: can this thing make money? Can it automate a factory? Can it write marketing copy? Can it handle customer service? The gap between Chinese models and top-tier US counterparts still exists on benchmarks, but Yao argues it's not the decisive factor for commercialization.
Why? Two reasons. First, Chinese companies can't access US models anyway due to restrictions. Second, Chinese users and enterprises care more about practical utility than theoretical capability. A model that scores 5% lower on MMLU but costs 80% less and runs locally is the better business choice.
From Free to Paid
The monetization shift is already happening. ByteDance introduced subscription tiers for its Doubao app in early May, ranging from 68 yuan (US$10) to 500 yuan per month. That's a big deal in a market where consumers historically resist paying for software. But Yao argues this reluctance is overstated — if the value is clear and demonstrable, Chinese users will pay.
The real battleground is enterprise. Consumer-facing AI features are nice, but the money is in transforming those features into reliable enterprise-grade infrastructure. Think: AI-powered supply chain management, automated quality control, intelligent customer service, predictive maintenance. The companies that can package AI into solutions that deliver measurable ROI are the ones that will survive the consolidation.
The Consolidation Is Coming
JPMorgan's view is that China's AI landscape is consolidating fast. The "hundred model" era — where every company built its own LLM — is unsustainable. Training and inference costs are too high, talent is too scarce, and the market can only support so many players. The winners will be those with:
1. Scale: Enough data and compute to train competitive models at reasonable cost
2. Distribution: Existing platforms and user bases to deploy AI features
3. Enterprise DNA: The ability to sell to businesses, not just consumers
That favors the big tech giants — Baidu, Alibaba, Tencent, ByteDance — but also leaves room for focused players like Moonshot (Kimi), DeepSeek, and Zhipu AI who have found specific niches or technical advantages.
🔥 Hot Takes
🔥 China's "good enough" AI is a feature, not a bug. The West obsesses over benchmark supremacy. China is building AI that works at the price point that makes business sense. In a world where inference costs matter more than training bragging rights, China's pragmatic approach might actually be the winning strategy.
🔥 The "hundred model war" was always a bubble. Did anyone really think China needed 100+ large language models? That phase was about capability demonstration and investor signaling. Now the real work begins: building products, finding customers, and making money. Most of those 100 models will be dead or acquired within 18 months.
🔥 JPMorgan's "master's-level graduate" standard is the new benchmark. Forget GPT-5.5 vs Claude Fable 5 vs Kimi K2.7 on arcane reasoning tests. The question is: can your AI do the work of a competent human employee? If yes, it's commercially viable. If no, it's a research project. This reframing changes everything about how we evaluate AI value.
🔥 ByteDance's Doubao pricing is the canary in the coal mine. When China's most aggressive growth-at-all-costs company starts charging for AI, the free era is over. If ByteDance can't subsidize AI indefinitely, nobody can. The subscription tiers — from $10 to $70/month — signal that Chinese AI is entering its commercial phase.
🔥 The US-China AI gap is becoming irrelevant. Yao explicitly says the performance gap doesn't matter for Chinese commercialization because Chinese companies can't access US models anyway. This means China is building a parallel AI ecosystem with its own standards, pricing, and use cases. The decoupling is real, and it's happening faster than most Western analysts expected.
The Bottom Line
China's AI industry is growing up. The wild-west phase of throwing money at ever-larger models is giving way to a more mature focus on enterprise value and sustainable business models. JPMorgan's analysis suggests this transition is happening faster than expected — and the companies that master the shift from "look at our cool model" to "here's your ROI" will define the next phase of Chinese AI.
The hundred model war is over. The enterprise value war is just beginning.