- Price wars are common in China, now affecting AI chatbots.
- China’s AI landscape shifted from lacking large language models (LLMs) to rapidly advancing, narrowing the gap with the U.S.
- Chinese LLMs now rank among the world’s top 20, with some outperforming their Western counterparts.
- Rapid development of over 100 LLMs, many with more than a billion parameters, driven by potential profits.
- Price competition is intense due to lack of a clear technological leader in China, unlike in the U.S.
- Significant price cuts by major players like High-Flyer, ByteDance, Alibaba, Baidu, and Tencent.
- Concerns over “war of a hundred models” reducing revenues and investment in necessary computing resources.
- American sanctions limit Chinese access to advanced AI chips, exacerbating competitive challenges.
- Likely outcome is market consolidation by giants such as Alibaba, Baidu, ByteDance, and Tencent.
- Smaller companies may be squeezed out or seek opportunities in international markets.
- Potential for many Chinese LLMs to fail sooner than expected due to intense domestic competition.
Main AI News:
Price wars are ubiquitous in China. The swift rise of hundreds of similar companies is driving down retail prices of everything from electric vehicles to bike-sharing and bubble tea. The latest entrants to this cutthroat competition are artificial-intelligence chatbots. This may seem unexpected. Until recently, China’s issue was not an excess of large language models, the type that enables ChatGPT to create humanlike content, but their scarcity. At the start of 2023, experts believed that existing Chinese LLMs lagged a decade behind the American frontier.
Now, the same experts believe that the transpacific model gap has narrowed to a year or less. LMSYS Chatbot Arena, which ranks LLMs’ performance, lists five Chinese ones among the world’s top 20. On some metrics, a model named SenseNova 5.0, developed by SenseTime, a Shanghai-based AI firm, surpasses the LLM that powered ChatGPT in logical reasoning and creative writing before its May upgrade. In February, market-research firm AskCI predicted that Chinese LLMs will generate 22 billion yuan ($3 billion) in revenue this year, up from 15 billion yuan in 2023. By 2028, they could be generating five times as much.
These impressive technological advancements and the potential for wealth have encouraged numerous AI hopefuls to join in. Last year, Robin Li, CEO of Baidu, a Chinese search giant with AI ambitions, noted that China was producing one LLM a day. While he may have exaggerated for effect, he was not far off. According to government estimates, the country now has more than 100 models with over a billion parameters, roughly as complex as some versions of Meta’s popular Llama models.
Chinese AI developers are now rapidly turning these models into products. Unlike in America, where OpenAI has a clear technological lead, in China, users lack an industry benchmark to distinguish between the various products available, says Jeffrey Ding, a Chinese AI scholar at George Washington University. Unable to compete on technology, they are competing on price.
On May 6th, High-Flyer, a quant hedge fund with its own model, slashed prices for the latest version to one-hundredth of one yuan cent per 1,000 tokens (the preferred unit of LLM pricing). This is roughly 1% of what OpenAI charges for its premier model, GPT-4 Turbo. On May 15th, ByteDance, owner of TikTok and its Chinese counterpart Douyin, began offering its newest model at a similar discount. A week later, Alibaba, an e-commerce conglomerate, reduced prices for its flagship LLM by 97%. Hours later, Baidu announced that its Ernie chatbots would be free for all business users. Tencent, China’s largest internet firm, is also offering one of its LLMs for free.
A Tencent executive now refers to this as “the war of a hundred models”. Lee Kai-fu, a Taiwanese tech investor and author of the bestselling 2018 book “AI Superpowers”, has warned that this battle is a “lose-lose” scenario. Attracting more customers—and thereby accumulating more data to train ever smarter algorithms—matters little if the resulting lower revenues mean significantly less money to spend on the expensive computing resources required for training. This issue affects model-makers globally but is particularly severe for Chinese ones. They are barred by American sanctions from accessing American-made AI chips, which are the most powerful and efficient worldwide.
In the long term, the most likely outcome of the price war is the consolidation of the Chinese AI industry into the hands of a few wealthy digital giants like Alibaba, Baidu, ByteDance, and Tencent. Promising smaller model-makers such as Baichuan, Moonshot, and Zhipu AI might be pushed out of the domestic market—despite the fact that, as OpenAI and other upstarts like Anthropic and Mistral have shown in the West, newcomers have a record of out-innovating incumbents in machine learning.
Some Chinese challengers may attempt to try their luck in international markets. 01AI, Mr. Lee’s latest venture, has indicated that if the price war becomes uncontrollable, it is ready to look abroad for growth. However, many domestic rivals would find it difficult to navigate foreign technology markets and thus remain trapped in their highly competitive home market. A hundred LLMs may be flourishing in China. Many of them may wither much sooner than their creators anticipated.
Conclusion:
The AI chatbot price war in China signifies a highly competitive and rapidly evolving market. Major tech giants are leveraging aggressive pricing strategies to capture market share, which could lead to the consolidation of the industry among a few dominant players. However, the intense price competition risks reducing overall revenue and investment in necessary computing resources, potentially stalling further innovation. The challenge is compounded by restrictions on access to advanced AI chips due to American sanctions. Smaller companies may either be forced out of the domestic market or explore international opportunities, though many may not survive the current competitive pressures. This dynamic will shape the future landscape of China’s AI industry, potentially favoring large, well-funded corporations.
