China’s AI Model Race Is Just Getting More Crowded

Imagine Epic Games, Apple and DoorDash were developing their own large language models now from scratch to compete against the likes of OpenAI and Anthropic.

That’s basically what’s happening in China, where a trio of big names from other parts of the tech industry have jumped into an already-crowded race to make cutting-edge AI. MiHoYo, a Shanghai-based videogame maker behind such hits like Genshin Impact, is developing its own LLMs with the goal of becoming a top-tier lab in China in the next couple of years. Xiaomi, a leading smartphone maker that also sells electric cars, released its latest LLM two weeks ago. And Meituan, China’s biggest food delivery and local services app, in June unveiled a more powerful version of the LLM it launched last year.

The new entrants are set to further intensify the domestic AI race in China, where both tech incumbents and upstarts compete fiercely for talent and on pricing. That competitive cauldron has already produced models that have significantly narrowed the gap with the most advanced U.S. models, shaking up the market.

In the U.S., OpenAI and Anthropic have commanded the clear technical lead over a handful of followers, including Meta Platforms, SpaceX and Google. In China, no one appears to be giving up, yet no one has pulled meaningfully ahead of the others.

The three leading tech giants—ByteDance, Alibaba and Tencent—are still racing to develop their own models, bankrolled by profits from their existing businesses such as e-commerce and social media. Five other startups have been generating recurring revenue from their own models. Two of them, Zhipu (also known as Z.ai) and MiniMax, went public in Hong Kong in January; the other three—DeepSeek, Moonshot and StepFun—are preparing to go public soon. AI model leaderboards are often reshuffled whenever any of the eight companies release their latest offerings.

MiHoYo, which plans to invest up to 100 billion yuan ($14.9 billion) in AI development over the next three years, is trying to gain access to Nvidia’s advanced Blackwell chips to train more competitive models, just like leading Chinese AI firms do, according to a person with direct knowledge of the efforts. Co-founder Cai Haoyu, the company’s largest shareholder and the mastermind behind its blockbuster games, is overseeing the foundation model team, according to the person.

Xiaomi founder and CEO Lei Jun, meanwhile, has talked publicly about integrating the company’s MiMo models into its gadgets and cars. The company made a splash in China’s AI community last year by poaching a star researcher from DeepSeek. Unlike Apple, which has partnered with Google to develop foundational models that power Apple Intelligence, Xiaomi is developing AI for its devices on its own. Xiaomi’s most recent flagship model—MiMo-V2.6-Pro, which was released last month—is ranked ninth on the Artificial Analysis Intelligence Index, on par with xAI’s Grok 4.7 and ahead of the flagship models from Alibaba and Z.ai that were launched several weeks ago.

Xiaomi declined to comment. MiHoYo and Meituan didn’t reply to a request for comment.

The Chinese AI model landscape is less crowded now than in 2023, when countless Chinese startups took inspiration from ChatGPT and rushed in to develop LLMs in what local media described as the “war of 100 models.” But the consolidation many predicted back then hasn’t happened yet.

China’s tech companies generally deeply distrust each other, and executives view it as more cost-effective to build their own AI than to buy competitors. For example, Meituan and Alibaba compete on food and grocery delivery; Alibaba and ByteDance are both developing AI devices that could encroach on Xiaomi’s turf; miHoYo and Tencent are rivals in videogames.

Last year, Meituan co-founder and CEO Wang Xing said during an earnings call that the only strategy that makes sense for navigating the AI revolution is to “play offense” by investing in its own models—instead of playing defense to protect its existing business. Wang said that LongCat 2.0, the company’s latest model, which was released in June, gives the company a structural advantage in long-term cost control. LongCat 2.0 still trails behind many Chinese models.

“In China, founders oftentimes worry that not having their own models could leave them vulnerable in the future if they ever lose access to external platforms,” said Robin Zhu, senior analyst at Bernstein.

Another factor enabling so many companies to compete in AI is the widespread practice of distillation, the process of using outputs from frontier models like those by Anthropic and OpenAI to train their own models, employees of Chinese AI labs say. U.S. export controls on advanced chips to China have forced many Chinese labs to turn to distillation as a shortcut. While dependence on distillation could make it harder for the labs to differentiate, the availability of well-established techniques and resources for distillation also levels the playing field for all competitors in China.

The absence of mergers and acquisitions that drive consolidation is also helping sustain the crowded field. Eye-watering deals such as Advanced Micro Devices’ recent $8 billion acquisition of AI scientist Fei-Fei Li’s World Labs or SpaceX’s $60 billion takeover of Cursor are unheard of in China’s current AI boom.

While Alibaba and Tencent have made minority investments in many Chinese AI labs, they haven’t acquired them. “Chinese technology companies have historically been less willing than U.S. Big Tech to pay large strategic premiums to absorb adjacent startups or competing products,” said Rui Ma, an angel investor and founder of the newsletter Tech Buzz China.

Whereas acquisitions are a normal growth strategy in Silicon Valley, often rewarded by stock market investors, “China developed much more of an IPO-or-nothing mentality,” Ma said. “That makes it difficult for an acquirer to offer a price compelling enough for founders and investors to sell, and it becomes self-reinforcing: if founders don’t expect to sell, buyers don’t develop the same acquisition and integration strategies either,” she said.

Jing Yang contributed to this article.

Juro Osawa is a reporter covering tech in Asia, from Alibaba and Tencent to startups. He previously worked for The Wall Street Journal. He is based in Hong Kong and can be found on Twitter at @JuroOsawa.