China now has more than 200 large language models (LLMs), according to the state-run Global Times, with the country accounting for 40% of the global total, just behind the United States' 50% share, according to brokerage CLSA.
The phrase "war of a hundred models" was coined by a senior Tencent executive to describe a frenzy of AI development triggered by the success of OpenAI's ChatGPT, and name has stuck.
Shakeout begins
JPMorgan's head of China equity research, Alex Yao, told the South China Morning Post that the hundred-model war is gradually coming to an end, with the market now consolidating around players capable of converting consumer AI products into corporate infrastructure.
"Only those with the strongest capabilities will survive," said Esme Pau, head of China internet and digital asset research at Macquarie Group, who forecasts a period of consolidation followed by a price war as companies fight for users.
Revolution or involution?
There are signs the process has begun, with Beijing simultaneously, yes, concerned, but also possibly inflaming the problem.
The government has warned tech firms against "involution", a term used to describe races to the bottom that have already driven solar and EV prices into the ground.
But at the same time, it is subsidising the AI industry's compute and energy costs, a contradiction Semafor noted keeps unprofitable companies solvent and the spiral going.
It's China's version of circular investment we are seeing in Silicon Valley, which could have far-reaching implications for the global economy.
Anyway, back to the pricing problem that is troubling the authorities in the People's Republic.
ByteDance kicked off the first round of AI discounting by pricing access to its LLMs at 99.8% below GPT-4, with Alibaba cutting its own model prices by 97% in response and others following.
Bank of America Securities analysts attributed the dynamic to narrowing capability differences, pointing to "limited capability gaps across incumbents" as the reason price has become the only remaining lever, compressing margins and pushing model APIs toward commodity status.
"The price war is reshaping how AI is accessed in China," one analyst told Roic.ai, speaking anonymously due to the sensitivity of ongoing negotiations.
Going global
Outside China, the effects of that competition are showing up in usage data.
Chinese open-source models processed a weekly peak of 46% of enterprise API tokens on OpenRouter by mid-July 2026, according to Digital Applied's Q2 2026 market share report, having held above 30% every week since February.
Alibaba's Qwen family has overtaken Meta's Llama in cumulative downloads on Hugging Face.
And one documented enterprise case found fine-tuning Qwen for a specific application delivered roughly 30% cost savings against an equivalent Western-model setup.
By Q2 2026, ten providers carried the meaningful share of Chinese AI output: Xiaomi, Alibaba, Zhipu, DeepSeek, Moonshot, MiniMax, StepFun, ByteDance, Baidu and Tencent, with Baichuan, Yi, Xunfei and KwaiKAT operating as second-tier niche players.
Among the established names, Tencent remains the laggard, though it has hired a former OpenAI researcher to lead its AI efforts. ByteDance holds a global leadership position in video generation. Alibaba dominates open-source.
IPO wave
The shakeout is now being channelled through the public markets.
Zhipu AI and MiniMax made back-to-back debuts on the Hong Kong Stock Exchange in January 2026, becoming the first pure LLM makers to list in the financial hub. MiniMax raised $620 million on debut and doubled its share price on the first day of trading.
Moonshot, the developer of the Kimi chatbot, has since filed confidentially for a Hong Kong IPO targeting a $3 billion raise, at a reported valuation of $50 billion, with Goldman Sachs, CICC and Deutsche Bank working on the deal. Its annual recurring revenue is expected to reach $2 billion by December, up from $1 billion currently.
StepFun is also preparing to file for a Hong Kong listing, with investors proposing a valuation of up to $12 billion.
Pivot or perish
Baichuan is taking another route. It is narrowing its focus from general-purpose AI to healthcare.
The pivot drags it out of the mass market and gives it a fighting chance of finding a defensible niche.
Is this the next phase? Pivot or perish?
