Rise of the Chinese AI dragons

The advent of transformer-based models has revolutionized the field of artificial intelligence, ushering in a new era of powerful and efficient AI systems. Transformers, introduced in 2017, have become the backbone of many modern AI applications, including natural language processing (NLP) and computer vision. These models leverage self-attention mechanisms to process data more effectively, enabling them to handle complex tasks such as language translation, text summarization, and image recognition with remarkable accuracy.

China has been at the forefront of adopting and advancing transformer technology. The country’s robust academic and research infrastructure, combined with significant government support, has fostered a fertile environment for innovation. Chinese universities and research institutions have made substantial contributions to the development of transformer models, publishing numerous influential papers and developing cutting-edge techniques.

The race for AI supremacy has intensified in recent years as nations realized its strategic importance. AI has the potential to transform entire industries, shortening product development cycles, and automating knowledge work at scale. The United States, China, and other leading economies are investing heavily in AI research and development, aiming to secure a dominant position in this rapidly evolving field.

Big Tech to Big AI
Most of technology in the non-China world is led by BigTech (The Magnificent Seven)- western technology behemoths from Google to Microsoft to Nvidia. Other dominant players have also evolved, notably OpenAI which led the breakthrough with its ubiquitous ChatGPT.
OpenAI, backed by Microsoft, continues to set new benchmarks with newer large-language-models (LLMs). Microsoft has been quick to integrate the nascent AI technologies into into its Office productivity suite, Github and Azure Cloud with CoPilot. Google’s Gemini project aims to develop multimodal AI models capable of handling a wide range of tasks. Anthropic’s Claude and Mistral’s Le Chat are also making waves with their innovative approaches to AI development.

Most of the underlying research of these corporation rely on leveraging advanced chips technology (read H100 and more advanced GPUs from Nvidia) to train their models. Training advanced LLMs now run easily into millions of dollars. However, the emergence of cost-effective alternatives from China is beginning to challenge their dominance.
As early as mid-January, Chinese startup DeekSeek launched its free AI assistant that it claimed uses less data at a fraction of the cost of its Western incumbents.

Protectionist US tries to hobble China
Since ChatGPT went viral and triggered the GenAI race, the demand for Nvidia GPU chips, which was already in short-supply, skyrocketed. Prices went up, as did Nvidia’s market cap.

• In addition, the US government passed its CHIPS Act in 2022, providing massive funding to expand semiconductor fabs and chip production domestically, while prohibiting recipients from expanding semiconductor manufacturing in China and countries deemed to be security threats to the US.
• New export controls specifically targeting China were implemented, targeted to restrict China’s ability to obtain advanced computing chips
• As early as 2025, the US further tightened controls on the global AI supply chain^.

Paradoxically, such sanctions seem only to have spurred China to supercharge its chip industry, growing faster than anywhere else in the world.

DeepSeek upheaval
DeepSeek made headlines with its highly cost-efficient AI model, which sent shockwaves through the market. DeepSeek’s model, built at a fraction of the cost of leading U.S. models, signals the potential for a new price war in AI.

Unlike models from OpenAI and Meta, which are trained on expensive high-end semiconductors, DeepSeek’s model is allegedly 45 times more efficient. This efficiency has raised concerns about the future of high-priced AI chips, particularly those from Nvidia. Moreover DeepSeek open-sourced its advanced reasoning model R1 and threw in a free chat assistant to boot.

Nvidia, the dominant player in AI chip design, saw its stock price tumble after DeepSeek’s latest model demonstrated its efficiency. The potential for more affordable AI models driven by China’s advancements could accelerate the adoption of AI technologies globally, but it also raises questions about the sustainability of high profit margins in the AI industry. DeepSeek’s rise underscores China’s growing prowess in AI and its ability to challenge Western dominance in this critical technology.

The 6 AI Dragons of China
China’s AI ecosystem has given rise to a group of standout startups known as the Six AI Tigers/Dragons (pick your favorite mascot)." These companies: Stepfun, Zhipu AI, Minimax, Moonshot AI, 01.AI, and Baichuan AI - are at the forefront of China’s AI sector, each bringing unique strengths and capabilities.

1. Stepfun, founded by former Microsoft senior vice president Jiang Daxin, has quickly become a contender with its portfolio of foundational models.
2. Zhipu AI, incubated by Tsinghua University in 2019, debuted China’s first self-developed pre-trained large language model (LLM) in August 2022. It’s also China’s biggest AI startup by number of employees (~800).
3. Minimax was setup by Yan Junjie, who previously led deep learning toolchain and algorithm development at SenseTime. It has found success with its AI platform Glow and its overseas app Talkie.
4. Moonshot AI, founded by Yang Zhilin is a well-known figure in the field of natural language processing (NLP). He previously worked at Google AI and was involved in the development of some of the leading language models.
5. 01.AI, a newcomer was founded in 2023 by Kai-Fu Lee, an AI expert and prominent investor who helped Google and Microsoft get established in China. It had a $200 million funding from Alibaba and others. Lee, in his provocative 2018 book AI Superpowers, had argued that Chinese AI labs and companies would soon rival those in the US thanks to the country’s abundance of talent, data, and users. This is exactly how it seems to be unfolding.
6. Baichuan AI was setup by Wang Xiaochuan, a well-known figure in the Chinese tech industry, formerly the CEO of Sogou, a search engine and internet services company. are also making significant strides, focusing on different aspects of AI applications and development. The company is backed by Alibaba, Xiaomi, Tencent, and Chinese state-backed AI funds.

Apart from these six, Chinese ecommerce major Alibaba group’s Qwen and relative newcomer DeepSeek are also at the forefront of AI research and development. Deepseek has been setup by the quant hedge fund High-Flyer co-founder Liang Wenfeng.

Here’s a summary table of the notable AI leaders of China:

China’s rapid progress in AI is reshaping the global technological landscape.
The country’s focus on developing low-cost, powerful AI models is not only driving innovation but also making AI more accessible to a broader range of users. The rise of startups like DeepSeek and the Six AI Tigers demonstrates China’s commitment to achieving AI leadership. With increasing geopolitical tensions, the race for AI supremacy is only going to get closer than ever. China’s deep talent pool of researchers and scientists as well as strengths in scaling low-cost industrial production will make sure that it playsa pivotal role in shaping the future of this transformative technology.

References:

• US export controls on chips manufacturing items to China
• ^US controls on global AI chip flows
• Alibaba QwenLM | Qwen Chat
• DeepSeek R1 reasoning model | DeepSeek chat
• Moonshot.ai Kimi chatbot
• Wired profile on O1.ai