Mar 15, 2023
China Can Quickly Catch Up to US AI, Says Venture Veteran
(Bloomberg) -- China can match the US in artificial intelligence thanks to the expertise of companies from Alibaba to Baidu, joining a global tech transformation that will dwarf the mobile revolution, according to industry pioneer Kai-Fu Lee.
American companies like Microsoft Corp. and Alphabet Inc. have the clear lead now, but much as it did in the early days of the internet, China will catch up through quick iteration from its private sector, said Lee. The bestselling author on AI, who founded Sinovation Ventures more than a decade ago, has been at the heart of China’s growing adoption of the technology and backed some of its biggest startups in the field like Megvii and Meitu.
Firms around the world are rushing to show off their latest AI creations since OpenAI Inc.’s ChatGPT demonstrated the technology’s potential to a wider audience. Baidu Inc. is set to unveil on Thursday its own ChatGPT rival that will be China’s most significant entry in the race with the US to stake out the nascent arena. Lee called the Beijing firm China’s obvious frontrunner in this field, though it will be far from alone.
“The US continues to be the world’s breakthrough innovator,” the 61-year-old VC founder said from his office in Beijing in his first interview with foreign media since the end of Covid restrictions. But China’s internet giants like Alibaba Group Holding Ltd. and Tencent Holdings Ltd. are “all building large models on par with OpenAI, Microsoft, and Google.”
Lee’s 2018 book “AI Superpowers” was an early exploration of the rising conflict between the US and China in the technology industry, detailing the strengths and weaknesses of each country. He published “AI 2041” two years ago with science fiction writer Chen Qiufan to imagine what the technology was capable of decades into the future.
His optimism about China’s tech players comes despite intensifying trade sanctions by Washington cutting off access to the latest hardware and exacerbating divisions between the two internet spheres. The AI market will remain decoupled, he said, much in the way that internet services are today. But the former Apple Inc. and Microsoft executive sees that as just “small obstacles in light of a tidal wave revolution.”
US efforts to limit the rise of China as a geopolitical rival include orders to keep the gold-standard for AI-training semiconductors — Nvidia Corp.’s A100 and its successor H100 chip — out of China. Lee believes Chinese AI companies will be able to find substitutes, even if it won’t be an ideal situation for those seeking to do massive-scale training.
“There are approximate other chips that can be used as substitutes. Nvidia has the A800, which is pretty good,” Lee said, adding “it’s also not a desperate situation by any means.”
He also doesn’t foresee a complete divorce between researchers in the two countries, pointing to academic exchanges and sharing of best practices as means for collaboration at lower levels. “The researchers are still friends, they will borrow ideas from each other,” he said.
“It’s not Chinese or American. I think all the companies will tend to think of their core business and how they can integrate with AI,” Lee said. Comparing the integrated AI systems of the future to Microsoft’s Windows and Google’s Android, the VC chief said the AI 2.0 change will be a more radical transformation than the mobile era and vast prizes await the companies able to establish the definitive next-generation platforms.
“All these giants stand to gain a lot because the way in which this technology can connect to their existing product line and users is not difficult,” he said. “It’s in fact quite trivial.” Alibaba might see better click-throughs on its AI-enriched e-commerce outlets or ByteDance Ltd. could supercharge TikTok’s appeal with AI-generated video.
Market Mania for Everything AI Faces China, Google Reality Check
Still, the US retains a significant lead, in part because its companies got started earlier and in part because they have better data with which to train their AI models — which determine exactly how useful a chatbot or content-generating tool will be.
“I’ve tried a few of these so-called Chinese ChatGPTs. I would say they’re not as good as ChatGPT,” Lee said. He pins the difference on the English-language services having more and higher-quality data than what is available to Chinese contenders. But “on the technologies, I was quite impressed by a few companies.”
It will take “maybe a year” for Chinese firms to get an equal or potentially larger amount of data, according to Lee. “China remains the hardest-working, most tenacious execution and value creator.”
As to who the eventual winners will be, Lee said startups looking to break through and outpace the bigger-budget incumbents must do two things.
“One is build everything AI-first, not mobility-first, not Windows-first, not-PC first. Secondly, leverage the power of open and the power of free,” he said, suggesting that open-source and free software has the chance for widest adoption. “But this is not for the ordinary startup, right? So we’ll have to see what company has that aspiration, experience and the guts to try to build something really, really big in an AI-first platform.”
--With assistance from Stephen Engle.
©2023 Bloomberg L.P.