What you need to know about Generative AI in China
If you follow me on social media, or are a fan of nexxworks’ monthly podcast Radar, you’ll know that I have become very passionate about Generative AI. As most of us over here tend to provide a very Western-centric vision about GenAI, I wanted to interview my friend, business Partner and China Keynote Speaker Pascal Coppens to offer us a fresh perspective from the East. Here are the highlights of our conversation:
The Big AI Plan
Before we dig deeper into the GenAI trend, let’s first circle back to the more general AI history of China. “It all started in 2015, when president Xi Jinping announced that China wanted to become the global innovation leader”, explained Pascal. “Shortly after, in 2017, China was one of the very first countries to introduce a national AI plan. It wanted to become an AI superpower, not just because it felt that was important for society, but because this technology would ultimately allow it to grow its economy and its GDP (which had been slowing down growth-wise).”
The AI plan turned out to be very successful, especially in the consumer area, according to Pascal. AI first company Ant Financial, for instance, became the biggest FinTech in the world. Ping An became one of the biggest players in the insurance industry. Chinese electrical vehicles - with a lot of embedded AI to make them autonomous - are now flooding our market. A lot of China’s AI ambitions were realized and Pascal believes that, with the current Generative AI movement, these wins might very well be dwarfed compared to what is being realized now.
The Great “Catch-Up”
When ChatGPT was launched about a year ago, on November 30th 2022, some people claimed that China had missed the boat on Generative AI. But they conveniently ignored that several big Chinese players like Baidu had already been working on large language models (the foundation of generative AI) of their own prior to that. These LLMs were being built, but they just weren’t available to the general public because the government had not yet allowed that. But that changed this summer.
Since then, Bytedance (parent company of TikTok), Baidu (the Chinese Google), Alibaba (the Chinese e-commerce giant), and many others were given permission to go public with their own LLMs. And, as is typical for China, this segment has been evolving like crazy over the past few months. Chinese companies are great at moving fast and scaling exponentially, as we know.
Since this summer, no less than 130 Chinese LLMs have surfaced, both from educational institutions (about 20) and private companies. Meanwhile, according to Pascal, it's estimated that 40% of all the large language models in the world were developed in China, 50% in the US and less than 10% in Europe.
So whereas, at the launch of ChatGPT, most educated Chinese had been using ChatGPT in English via VPNs (virtual private networks), more and more of them are starting to use the local LLMs, which have been trained in their own language.
A Cold War on Talent
The US got ahead in the first Cold War because of two specific measures. First, they invested a massive amount into national security technology, which led them to win the space race. Secondly, they made a huge investment in talent, with the National Defense Education Act of 1958. As a result, college attendance - in the preferred study fields of science, technology, engineering, mathematics and languages - jumped from 5% to 15% in just a few years’ time.
My feeling is that, in this current Cold War between China and the US, the latter is making a lot of investments, but they are neglecting the talent part. While China, on the other hand, has a very strong focus on that. Pascal for instance explained that China’s big AI plan has a specific focus on talent and education. Universities like Tsinghua University, Beijing University and Fudan University in Shanghai, are set on educating more AI talent than ever before. While local governments and local research institutions in Shanghai, Guangzhou, Shenzhen and other cities are building R&D hubs in order to foster talent.
On top of that, clarified Pascal, the Chinese are really pragmatic. Seeing that AI expert is one of the best paying job these days, Chinese parents are incentivizing their children to study that specific field in order to safeguard their future. Also, many Chinese people are studying AI because it will allow them to gain respect from the community as well as visibility, which is hugely important over there.
It is also interesting to note that, whereas many young Chinese used to study and then stay abroad, the tech war between China and the US has changed that. According to Pascal, more and more experts are returning to China after their studies, simply because it's much easier for them to build a reliable future over there. It’s not just about the available infrastructure, funding and market demand. It’s also about often feeling disrespected or treated like spies abroad.
The Chinese GenAI ecosystem
Pascal also pointed out an interesting difference between the American and Chinese GenAI ecosystem. In the west we saw this fast evolution of relatively small startups like OpenAI and Anthropic into big players. The traditional big tech companies like Microsoft, Amazon and Google have been mostly participating by proxy. Google acquired DeepMind in 2014 and that’s where a lot of their AI effort has originated. Microsoft has a very strange relationship with OpenAI, where the latter is doing the “cool stuff” which Microsoft then puts it into their products. We saw a very similar move with Amazon AWS and Anthropic.
In China, it’s almost the opposite: the Big Tech companies are leading the way, with Baidu as the first mover. The reason is actually quite simple. You need a lot of chips, performance, computing power and talent to build GenAI. Big Tech - the Baidus, the Alibabas and the Tencents (called the BAT) - had the money for that. Startups did not. It was still quite difficult for the latter to get the necessary funding because of the resonating consequences of the pandemic. So the traditional tech players, who had the computing power, were the ones in pole position to actually start developing GenAI in China.
But they did not do it alone. They collaborated with universities and startups, often for industry specific solutions. Alibaba, for instance, has funded quite a few newcomers, like AI startup Baichuan. That’s actually a special one, because Alibaba’s biggest competitor Tencent (and even Xiaomi) invested in it as well. So you know it’s definitely worth to keep an eye on. The same goes for Zhipu, from Tsinghua University, which is also funded by both Alibaba and Tencent. A third special one was founded by Kai-Fu Lee, the author of bestseller “AI Superpowers”: a company called 01.AI which wants to develop a homegrown large language model for the Chinese market. Alibaba also invested in this one, and it has grown into a unicorn in just a couple of months’ time.
The AI service leader of the world
One of the most important changes that the GenAI trend will entail, according to Pascal, is that China will evolve from a production-driven background - the “factory of the world” - to a service-driven industry with generative ai.
“China is set on replacing the West as the leader of the service industry worldwide”, said Pascal. “Currently, 78% of US GDP is derived from the service industry. China used to pale compared to that. In the 1980s, services accounted for a mere 22% of China’s GDP. Today that number has grown to 54% and they want to increase it even further. The current geopolitical war is not just about chips or technology. It's about brains, knowledge, insights and content and GenAI may very well come to play a pivotal role here in the future. The Chinese believe that, if their service industry grows, this will also increase consumption, as well as jobs. It will boost innovation, productivity and living standards. China is expecting a lot from the GenAI service movement.”
Hauwei’s Big Ambitions
A perfect example of this transition to the service industry is Huawei, said Pascal, which over the years evolved from a traditional telecom device company to a software and service company.
I've always seen Huawei as playing in the same space as the Alcatels and the Nokias of this world. But most of those companies have stuck to bare metal, to hardware. We have not seen any of them move into the realm of cloud. While that is exactly what Huawei did, full speed ahead. 80% of the Chinese cloud is in fact in Alibaba’s hands, which makes it a lot like Amazon over here. But Huawei’s cloud targets more vertical solutions, for instance for airports or the government.
Huawei is investing a lot in this transition. In September, at Huawei Connect, CFO Sabrina Meng stated that that over the next 10 years her company will fully focus on intelligence and AI. They have developed generative AI solutions for government services, for the railroad sector, for weather forecasting, for mining, etc. They are really concentrated on building vertical GenAI applications to help improve the CX as well as the productivity of their customers.
The irony is that Huawei had no other choice than to pivot to several new domains like the cloud or even chip making. They were indirectly forced by the US, because they no longer had access to essential technologies. They couldn't use Android for their phones anymore. They couldn't use chips for their smartphones and infrastructure. And so they were simply forced to build their own, which allowed them to become even more advanced and powerful. Which is exactly the opposite of what the US had in mind with these tech bans.
Pascal really believes that we’ll see this transition - from the factory of the world to the service leader of the world - reflected in many more companies in the coming months and years and that GenAI will play an important part here. It’s going to be a really big change, but if China succeeds, we’re going to experience very interesting times.
Pascal and I talked about so much more in our podcast conversation in a special 'Radar - by nexxworks' episode - from AI regulation to education to the cloud and light weight LLMs or geopolitics - so be sure to check the full conversation.
You can also watch the video version:
This was an instalment from Peter's newsletter.
Read this and more here.