Chinese AI Startups Make Gains in Challenge to U.S.-based OpenAI (AI Translation)
Listen to the full version

文|财新周刊 刘沛林
By Liu Peilin, Caixin Weekly
2024年是生成式人工智能(AI)飞速发展的一年。时至年末,经过一年多战算力、战模型、战融资的“百模大战”,中国六家估值超过10亿美元的“独角兽”AI创业公司,被市场视为“大模型六小虎”头部阵营,暂时稳住阵脚。
The year 2024 marks a period of rapid advancement for generative artificial intelligence (AI). By the year's end, following over a year of intense competition in computing power, model development, and fundraising in what’s referred to as the "Hundred Model Battle," six Chinese AI startups, each valued at over $1 billion, have emerged as market leaders. Dubbed the "Big Model Six Tiger Cubs," these companies have secured their positions in the industry.
The "Six Little Tigers" are: Zhipu AI, MiniMax, Baichuan Intelligence, Dark Side of the Moon, Jie Yue Xing Chen, and ZeroOne Universe, with the first four companies currently valued at over 20 billion yuan.
“六小虎”之所以受资本青睐,是团队背景、模型能力、产品形态等各要素的综合结果。六家公司创始人近期分别接受财新采访,在他们看来,中国的AI大模型在某些指标上已基本拉平和领先者美国OpenAI公司的距离。
The "Six Little Tigers" are favored by capital due to a combination of factors including team background, model capabilities, and product forms. The founders of these six companies recently spoke with Caixin, expressing that in their view, China's AI large language models have on some metrics nearly matched the distance with the leading U.S. company OpenAI.
- DIGEST HUB
- In 2024, six Chinese AI startups, valued over $1 billion, emerged as leaders in the "Hundred Model Battle," advancing generative AI technology significantly. These "Six Little Tigers" include Zhipu AI, MiniMax, Baichuan Intelligence, among others.
- These startups face challenges from restricted U.S. chip imports and competition with domestic internet giants. They focus on algorithm optimization and differentiated applications to keep pace with global leaders.
- While China's AI startups lag behind U.S. firms like OpenAI in terms of computing power and finance, they continue to innovate with government and strategic investments to close the technological gap.
The year 2024 was significant for generative AI, with the emergence of six Chinese AI startups, collectively known as the "Big Model Six Tiger Cubs," as market leaders. These companies, comprising Zhipu AI, MiniMax, Baichuan Intelligence, Dark Side of the Moon, Jie Yue Xing Chen, and ZeroOne Universe, each have a valuation exceeding $1 billion, stemming from intense competition in computing power and model development. [para. 1]
These firms have gained favor with investors due to their team backgrounds, model capabilities, and product offerings. The rapid growth of these startups, incubated amid China's strategic push in AI, is driven by differentiated approaches to computing power and application development, particularly in contrast to established domestic internet giants. Zhipu AI, in particular, has emerged as a frontrunner, being the earliest founded among the "Six Little Tigers" and the first in China to release a self-developed pretrained language model. [para. 2][para. 4]
Strategic differentiation is key to surviving intense competition. Startups must navigate restrictions on advanced computing chip imports from the US while leveraging China's vast AI talent pool to optimize algorithms. Additionally, they face established giants like ByteDance, which have significant traffic and application development advantages. [para. 3]
The development trajectory of these startups varies. Zhipu AI has cultivated a user base of millions through enterprise and consumer-focused deployments, while MiniMax's pivot from SenseTime led to the launch of several successful AI-based consumer products like Talkie, which experienced substantial adoption in the US. [para. 5][para. 7]
Baichuan Intelligence focuses on large model applications in industries such as healthcare, finance, and education, securing significant valuations and partnerships within a short timeframe. Their approach exemplifies the startups' strategy to apply large models broadly across sectors while steadily building market penetration. [para. 9]
Meanwhile, companies like Dark Side of the Moon capitalize on high-profile academic and technical leadership to delve into long-text features and distinct consumer applications. This company, among others, demonstrates the careful choreographing of technical talent and distinctive product niches as pillars for long-term success. Their philosophy prioritizes maintaining a lean team to enhance innovation, avoiding the pitfalls of excessive scaling experienced by prior "Four Little Dragons." [para. 14]
In parallel, companies like ZeroOne Universe, founded by renowned investor Lee Kai-Fu, aim to establish comprehensive AI ecosystems. This highlights a holistic approach to AI infrastructure development tailored to constraints in computational resources faced domestically compared to counterparts like OpenAI with vast computing clusters. [para. 16]
Despite current differences in computing power and financial backing compared to US companies, the Chinese startups recognize the potential in alignment with government-backed infrastructure projects and strategic partnerships. They face resource constraints but often pursue applications for immediate revenue while slowly advancing toward AGI development, promising cost efficiencies and localized solutions. [para. 24][para. 27]
Future growth hinges on overcoming funding hurdles, with Zhipu AI and others securing successive financing rounds from both venture and state-affiliated entities. A sustainable path for these companies involves forging alliances with government and large enterprises to capture and innovate in application markets. This highlights the vital role of government-supported initiatives in securing computing resources and fostering industry resilience, underscoring the unique environment influencing the trajectory of China's AI scene. [para. 32][para. 36]
- Zhipu AI
智谱AI - Zhipu AI, founded in 2019 and incubated by Tsinghua University, is one of China's top AI startups, known as the "Big Model Six Tigers." Led by Zhang Peng and Tang Jie from Tsinghua's Computer Science Department, it's China's first startup to release a self-developed pre-trained language model. With nearly 1,000 employees, its app “Zhipu Qingyan” boasts 25 million users and annual revenue over ten million yuan. The company aims to be China's OpenAI.
- MiniMax
MiniMax - MiniMax, founded by Yan Junjie in December 2021, gained attention with its AI virtual social product Glow and later Talkie, which became a top entertainment app in the US. The company introduced the productivity tool Heluo AI in April 2024, achieving nearly 10 million global monthly active users. MiniMax has around 400 team members and is valued at roughly $3 billion, with its main revenue coming from advertising and subscriptions.
- Baichuan Intelligence
百川智能 - Baichuan Intelligence was founded in March 2023 by Sogou's founder Wang Xiaochuan and former COO Ru Liyun. It focuses on AI model applications in healthcare, education, and finance. The company has around 400 employees and anticipates doubling by 2025. Baichuan has begun commercialization, expecting significant growth by 2025. The company's valuation exceeded $1 billion within six months of its founding.
- Moonshot AI
月之暗面 - Moonshot AI, founded by Yang Zhiyin, gained prominence in early 2024 with its long-text reading feature, boosting user base significantly. The company is known for its efficient use of resources, maintaining a smaller team while achieving a high user-to-GPU resource ratio. The Kimi Chat platform has 36 million monthly active users. Moonshot AI focuses on innovation and careful resource allocation, leveraging its long-text capabilities to stand out in the AI chatbot market.
- Transitional Starlight
阶跃星辰 - Transitional Starlight, founded in April 2023, is one of China's AI "Six Tigers." The company, led by former Microsoft VP Jiang Daxin, focuses on multi-modal output with products like "Yuewen" and "Maopaoya." It aims to establish a commercial ecosystem within 1-2 years with partnerships, including OPPO and Honor. The team consists of about 300 people and receives support from former Microsoft executive Harry Shum.
- 01.AI
零一万物 - 01.AI, founded by Kai-Fu Lee in March 2023, aims to create a comprehensive upper-layer application ecosystem by starting from foundational AI infrastructure. Lee emphasizes the need for AI infrastructure expertise for Chinese companies facing limited computing resources compared to Silicon Valley. They focus on cost, efficiency, and resource utilization, training models with significantly lower costs than OpenAI, while ensuring competitive performance both domestically and in international markets.
- SenseTime
商汤科技 - SenseTime, one of the "four AI dragons" known for visual AI, was previously involved with Junjie Yan, who later founded MiniMax. Yan had led general AI technology at SenseTime before starting MiniMax in December 2021, which quickly launched the AI virtual social product Glow.
- Sogou
搜狗 - Sogou founder Wang Xiaochuan, along with former COO Ru Liyun, founded Baichuan Intelligence in March 2023 during a new wave of large-model entrepreneurship following the success of ChatGPT. Baichuan focuses on applying large models in industries such as healthcare, education, and finance. Within six months of its establishment, Baichuan's valuation exceeded $1 billion.
- Tesla
特斯拉 - The article mentions Tesla CEO Elon Musk's AI company, xAI, which is building a 100,000-GPU compute cluster and has ambitious plans to expand its Memphis data center to house 1 million GPUs, with future goals reaching 10 billion GPUs. xAI has raised over $120 billion and, as of May 2024, has a valuation of $240 billion.
- OpenAI
OpenAI - The article highlights that OpenAI, a leading AI company from the USA, has achieved significant advancements with its models such as GPT-4o and GPT-o1. OpenAI's development includes a 10,000 GPU cluster for model training and a high valuation of $157 billion through substantial funding over $100 billion. The company is considered a benchmark for AI development globally, with Chinese companies striving to catch up with its technological progress.
- Amazon
亚马逊 - The article mentions Amazon in relation to Anthropic's collaboration, using tens of thousands of Amazon's self-developed AI training chips for training the next-generation model. Additionally, the context suggests that Amazon is involved in the competitive landscape of AI model development and deployment.
- Alibaba Cloud
阿里云 - Alibaba Cloud is actively promoting open-source large models. It has followed suit in adjusting its large model prices, affected by ByteDance's aggressive pricing strategies in the AIGC sector. The company is part of the competitive landscape where Chinese large model startups, traditional internet giants, and other market players are trying to find their footing amidst fierce competition and technology evolution.
- Tencent Cloud
腾讯云 - The article mentions that Tencent Cloud is among the companies that were forced to adjust their large model pricing strategy following the aggressive pricing set by ByteDance's cloud platform, Volcano Engine. This indicates Tencent Cloud's involvement in the competitive landscape of cloud services for AI models.
- Microsoft
微软 - The article mentions Jiang Daxin, the founder of Jiexue Xingchen, as a former Global Vice President at Microsoft and Chief Scientist at Microsoft Research Asia. Jiexue Xingchen has received support from Shen Xiangyang, Microsoft's former Global Executive Vice President, in the venture capital market.
- Google
谷歌 - The article mentions that Meta and Google are actively involved in open-source model development, which reduces the barriers for large model development. Google has also been working on open-source AI projects, alongside launching various applications to enhance AI capabilities and development accessibility.
- Meta
Meta - The article mentions that Meta is actively involved in open-source AI model development. Their efforts have helped lower the barriers to AI model development, making it more accessible. Additionally, Meta's Llama-3-405B model was trained on a large GPU cluster, emphasizing their significant investment in AI advancements.
- Xiaomi
小米 - The article mentions that Xiaomi is among several terminal manufacturers, including OPPO, vivo, Honor, and Li Auto, that are exploring the large model track either through self-research or collaboration.
- OPPO
OPPO - In the article, OPPO is mentioned as a collaborator with Jietuo Xingchen, one of the "six little tigers" AI startup companies. They are working together to become AI assistants on the mobile end.
- vivo
vivo - The article mentions that vivo, along with Xiaomi, OPPO, Honor, and Li Auto, is entering the large-scale AI model race through self-development or partnerships, amidst the rise of generative AI in China.
- Honor
荣耀 - Honor has partnered with Jizepu AI to integrate AI capabilities into mobile devices. This collaboration aims to develop AI assistants for mobile platforms, utilizing Jizepu AI's advanced models. This involves implementing stronger intelligent agents, enabling complex operations across different devices and applications by leveraging enhanced foundational models and family business logic to offer a seamless AI user experience.
- Li Auto
理想汽车 - Li Auto is mentioned as one of the terminal manufacturers in China exploring the large model AI race, either through self-research or collaboration.
- NVIDIA
英伟达 - The article mentions that training large AI models typically requires thousands of NVIDIA GPU cards. However, due to U.S. restrictions on exporting NVIDIA AI chips to China, Chinese AI startups face significant challenges in accessing sufficient computational power. Despite these limitations, Chinese companies are using algorithmic optimizations to make do with fewer resources.
- Huawei
华为 - The article does not specifically mention Huawei. However, it briefly references Huawei in relation to domestic GPU resources, mentioning that Chinese companies may need to rely on larger scale Huawei or other domestic GPU resources for training large models due to limitations in accessing NVIDIA GPUs from the U.S.
- ByteDance
字节跳动 - In 2024, ByteDance launched the AI assistant "Doubao," gaining rapid market traction by leveraging its streaming platform, Douyin, for traffic and advertising. With competitive pricing, Doubao became the largest AI native app in China by user base, surpassing competitors like Kimi Chat and Baidu's AI assistants. ByteDance's powerful combination of technical and capital resources enabled substantial growth, making it a significant player in consumer AI applications.
- Meituan
美团 - The article does not provide specific information about Meituan. However, it mentions Meituan founder Wang Xing in the context of influential entrepreneurs in the tech industry comparable to ByteDance founder Zhang Yiming and Pinduoduo founder Huang Zheng, suggesting their potential involvement in the AI 2.0 era.
- Pinduoduo
拼多多 - The article does not mention Pinduoduo.
- Megvii
旷视科技 - The article mentions that team members from Moonlight were previously technical backbones at Megvii Technology. However, it does not provide any further details about Megvii itself.
- Samsung
三星 - According to the article, in November 2024, China's AI company Zhizhou AI signed a collaboration with terminal manufacturers, including Samsung, to deploy their intelligent agents on devices like PCs and smartphones. This collaboration aims to enhance AI-powered functionalities across various devices, reflecting a strategic move to embed AI capabilities within consumer and enterprise technology ecosystems.
- ASUS
华硕 - According to the article, ASUS signed a collaboration with Zhiyu AI to utilize its AI capabilities in their products. Zhiyu AI focuses on the development of intelligent agents and cross-platform applications.
- December 2021:
- Yan Junjie left SenseTime to found MiniMax
- August 2022:
- Zhipu became the first startup in China to release a self-developed pretrained language model
- March 2023:
- Sogou founder Wang Xiaochuan teamed up with former Sogou COO Ru Liyun to establish Baichuan Intelligence
- April 2023:
- LeapStar and Dark Side of the Moon were established
- May 2023:
- xAI had a valuation of $24 billion
- August 2023:
- MiniMax relaunched its product "Xingye" after obtaining necessary licenses
- March 2024:
- The updated Kimi Intelligent Assistant significantly enhanced its long-text reading capacity, leading to rapid adoption
- April 2024:
- MiniMax launched the productivity tool Hailuo AI
- May 2024:
- ByteDance's AI dialogue assistant "Doubao" was released
- September 2024:
- Wang Xiaochuan expressed that starting from 2023, China's AI startups had technologically caught up with expectations
- November 2024:
- Yang Zhilin reported Kimi Chat reached 36 million Monthly Active Users (MAU)
- December 2024:
- ByteDance introduced the Doubao Visual Understanding Model at a significantly reduced price
- MOST POPULAR



