Stay informed with free updates
Simply sign up to the Chinese business & finance myFT Digest — delivered directly to your inbox.
The writer is a founding partner of Hong Kong based VC firm IN. Capital
The two most important artificial intelligence ecosystems in the world are the US and China. Geopolitical tensions make it tempting to view the pair in opposition. Chinese AI has been described as both lagging and leading the US. In fact, the two countries are pursuing distinct strategies.
Admittedly, this is partly due to necessity. Since US start-up OpenAI launched chatbot ChatGPT in November 2022, the world has been consumed by a fevered race to build the large language models that fuel generative AI. China is handicapped in this race by two factors: lack of access to advanced US chips and censorship of information.
Without advanced chips, China’s compute power is constrained. Companies like OpenAI are expanding the size of their models. GPT-4, OpenAI’s latest flagship model, is reported to be trained on more than 1tn parameters. Chinese companies must instead focus on efficiency. Hangzhou-based DeepSeek, for example, released DeepSeek-V2 this year, an open-source large language model that uses 21bn active parameters.
Chinese companies are also hampered by the data they can access. The introduction of China’s Personal Information Protection Law in 2021 imposed tight data protection standards that are similar to Europe’s General Data Protection Regulation. In recent years, China’s regulatory bodies have repeatedly said that individual and corporate data need to be protected.
Such restraints mean Chinese companies have more incentives to produce AI services built on smaller large language models, instead of training them on the entire internet. They may lack the brute strength of their larger counterparts but they are cheaper to create and run.
China’s tech ecosystem is grounded in such pragmatism. Research that cannot be turned into a product within the foreseeable future, such as artificial general intelligence, struggles to attract funding.
This means that while the US ecosystem has the edge in groundbreaking innovations, China excels in execution: finding product-market fit, scale, and making applications highly affordable.
The World Intellectual Property Organization recently published a patent landscape report on generative AI that showed China was responsible for more than 38,000 filings between 2014 and 2023, compared with 6,276 by the US over the same period. Tencent, Ping An Insurance, Baidu and Chinese Academy of Sciences led the pack. Of course, patents do not equal breakthroughs. Alphabet ranks below Tencent in terms of AI patent titles but Alphabet’s Google DeepMind gave the world AlphaFold, which could be transformative to AI drug discovery.
But the high volume of China’s patents may translate to more products. One area of Chinese AI application to watch is electric vehicles. China spent 26 years producing its first 10mn EVs and only 17 months to produce the next 10mn. Roughly half of the cars sold in China this year are expected to be tablet-on-wheels smart cars. Baidu, which has just deployed 500 driverless taxis in Wuhan, is conducting the world’s largest autonomous vehicle experiment. The companies may never produce superhuman AI but “good enough” AI interacting with humans via their cars will be popularised in China before anywhere else.
China’s tech sector is often characterised as a monolithic goliath underpinned by state-driven ambition. The reality is that regulation and geopolitical restrictions mean it has the edge in some areas and is struggling to catch up in others. AI is the most transformative technology in the world. Both China and the US have contributions to make to it. After all, the most profound challenges our species face — from fighting climate change to curing cancer — are not Chinese or American but human.
Read the full article here