In AI rivalry with the U.S., China is behind on a key asset: Its own OpenAI

Nvidia’s rocket-ship ride in the stock market underscores the extent to which chip quality and availability will dictate the winners in the generative AI era. But there’s another aspect to measuring early leads in the space. In China, which is angling to produce its own chips or get more from Nvidia, no dominant gen AI contender to OpenAI has emerged yet among dozens of Chinese tech titans and startups.    

Late to the game, China is seeking to catch the lead of OpenAI in a wider U.S. AI market shaped by tech titans Microsoft, Alphabet’s Google and Amazon, and well-financed startups including Anthropic, which this week received a $2.7 billion infusion of cash from Amazon.

In the fast-moving field, the gap between the U.S. and its tech rival China is seen as wide. “The leading Chinese companies are benchmarking against ChatGPT, which indicates how far behind they are,” said Paul Triolo, senior vice president for China and technology policy lead at Dentons Global Advisors in Washington, D.C.

“Not too many companies can support their own large language model. It takes a lot of capital. Silicon Valley is definitely well ahead of the game,” said Jenny Xiao, a partner at AI VC firm Leonis Capital in San Francisco.

The U.S. remains the biggest investment market. Last year, funding of gen AI upstarts accounted for nearly half of $42.5 billion invested globally in artificial intelligence companies, according to CB Insights. In the U.S., VCs and corporate investors drove AI investment to $31 billion across 1,151 deals, led by large outlays in OpenAI, Anthropic and Inflection. This compares with $2 billion in 68 deals in China, which marked a large drop from 2022’s $5.5 billion in 377 deals. The fall-off is partly attributable to restrictions on of U.S. venture investment into China.

“China is at a big disadvantage in building the foundation models for Gen AI,” said Rui Ma, an AI investor and co-founder of investment syndicate and podcast TechBuzz China.  

But where China lags in foundational models, which are dominated by OpenAI and Google’s Gemini, it’s closing the gap by using Meta’s open source, large language model Llama 1, and Triolo said the Chinese contenders, if behind, are improving on the U.S. model.

“Many of the China models are effectively forks of Llama, and the consensus is that these forks are one to two years behind the leading U.S companies OpenAI and its video-to-text model Sora,” Ma said.

China does have the tech talent to make a difference in the AI rivalry in the years ahead.

A new study by think tank Marco Polo, run by the Paulson Institute, shows that the U.S. is home to 60% of top AI institutions, and the U.S. remains by far the leading destination for elite AI talent at 57% of the total, compared with China at 12%. But the research finds that China leads the U.S. by a few other measures, including being ahead of the U.S. in producing top-tier AI researchers, based on undergraduate degrees, with China at 47% and the U.S. lagging with 18%. Additionally, among top-tier AI researchers working at U.S. institutions, 38% have China as their country of origin, compared with 37% from the U.S.

New Chinese gen AI market entries can also reach mass adoption quickly. Baidu’s ChatGPT competitor, Ernie Bot, released in August 2023, reached 100 million users by the end of the year. Samsung is planning to integrate Baidu’s Ernie AI into its new Galaxy S smartphones while in another high-profile development that speaks to U.S.-China relations, Apple is in talks with Baidu about supplying the iPhone 16 with the Chinese company’s gen AI technology. 

Within its current slate of AI contenders, Baidu’s Ernie Bot models are considered among the most advanced, according to Leong.

Several other Chinese companies are forging ahead, funded by major players in its own technology market. Large cloud companies such as such as Baidu and Alibaba, social media players ByteDance and Tencent, and tech companies SenseTime, iFlyTek, Megvii and Horizon Robotics, as well as research institutes, are all aiding the effort. 

Moonshot AI, funded by China’s e-commerce giant Alibaba and VC firm Hongshan (previously Sequoia China), is building large language models that can handle long content inputs. Meanwhile, former Google China president Kai-Fu Lee has developed an open source gen AI model, 01.AI, funded by Alibaba and his firm Sinovation Ventures.   

While China has accelerated development of its homegrown chip industry and advanced AI, its AI development has been limited in part by U.S. restrictions on exporting high-end AI chips, a market cornered by Nvidia, as part of a new battleground for tech supremacy between the U.S. and China. 

“Despite efforts to develop indigenous solutions, Chinese AI developers still largely rely on foreign hardware, particularly from U.S. companies, which is a vulnerability in the current geopolitical climate,” said Bernard Leong, founder and CEO of tech advisory Analyse Asia in Singapore.    

The ongoing tensions between the U.S. and China over technology innovation and national security issues is leading to a split in gen AI development, following the pattern of other impactful technologies caught up in superpower tech arms races. Given regulations and bans over sensitive, cutting-edge technologies, the likely outcome is two parallel ecosystems for gen AI, one in the U.S. and one in China. ChatGPT is blocked in China while Baidu’s Ernie Bot can only be accessed in the U.S. with a mainland Chinese cell phone number. “U.S. companies can’t go into China and Chinese companies can’t go into the U.S.,” Xiao said.  

U.S. Secretary of Commerce Gina Raimondo has stated that a goal of U.S. curbs on AI chip exports is to prevent China from acquiring or producing advanced chips. As mainland China focuses on homegrown capabilities, Chinese companies SMIC or Huawei could be an alternative to Nvidia. But the future for alternates is likely uncertain if export controls cut off these companies from the most advanced designs for manufacturing. Triolo noted that Huawei recently developed a series of AI chips as a rival to Nvidia.  

China is getting ahead in applying AI to certain categories, such as computer vision. “The chip shortage is very important for training foundational models where you need certain chips, but for applications, you don’t need that,” Ma said.

The “real killer app” for gen AI, according to Triolo, will be in companies that are willing to pay money to harness the technology as part of their business operations. Alibaba is focusing on integrating AI into its e-commerce ecosystem. Huawei, while competing more successfully against Apple’s iPhone in the consumer market in the past year, also has broader ambitions, developing AI for specific industries including mining, using its in-house hardware, Leong said.

Boston Consulting Group research suggests it may be a while before this wider gen AI market ramps outside of tech. Sixty percent of 1,400 executives surveyed are waiting to see how gen AI regulations develop, while only 6 percent of companies have trained their employees on gen AI tools.

AI and tech issues are front and center for China’s leadership, with the country’s release of guardrails on AI in 2023 after ChatGPT’s breakthrough, and then modifications of some measures.

The open source gen AI technology many Chinese developers use can encourage collaboration among globally and lead to shared insights as AI advances, but Leong said open source also leads to issues related to ensuring quality and security of the models, as well as managing bias and potential misuse of AI.

“China wants to make sure content is not spewing out. They also want their companies to lead and are willing to reign in draconian measures,” Triolo said.

Ethical and social concerns hinder gen AI advances in China as well as other regions, including the U.S., as see in the battle for control over OpenAI’s mission. Within China, there is another factor that could slow AI acceleration, according to Leong: maintaining control of gen AI applications, especially in areas sensitive to state interests.