In a statement released by American artificial intelligence company Scale AI’s founder, Alexander Wang, it was claimed that the United States is ahead of Beijing in AI competition, with Chinese AI models essentially being replicas of American ones.
Scale AI primarily focuses on data labeling services. Over the span of eight years, the company’s market value has soared to approximately $14 billion, making it one of the unicorns in the American AI industry. Scale AI is closely associated with every major breakthrough in the field of AI and collaborates with the U.S. Department of Defense and global giant OpenAI.
According to Mr. Wang, a 27-year-old Chinese-American prodigy, if you observe the best AI models in China, they are essentially replicas of the American models. Therefore, he believes that the U.S. is the default leader in the field.
In the realm of AI, there are three fundamental pillars: computing power, algorithms, and data. In terms of computing power, no large-scale model can function without American chip company NVIDIA, and similarly, it cannot operate without Scale AI providing high-quality data labeling services. It has been likened to NVIDIA being the shovel seller of computing power, while Scale AI is the shovel seller of data.
As the founder and CEO of Scale AI, Wang warned that the U.S. needs to pay attention to how China is utilizing the internet and overseas data. From what is currently visible, the Chinese model is essentially replicating the American one.
Founder of Chinese mainland cybersecurity company Qihoo 360, Zhou Hongyi, admitted in an interview with the CCP mouthpiece “Global Times” that China is focusing on application development in the field of AI, and it may take one to two years for Chinese AI companies to catch up with the U.S. once American AI companies open-source their technology. However, there may still be barriers in computing power.
Mr. Wang stated that China has the concept of data factories, where they exploit the data of over a billion Chinese citizens to serve AI models, such as facial recognition and autonomous driving. While this gives China an advantage in AI development, their performance does not meet expectations.
“Now, even if China’s computing spending doubles, its effectiveness does not match the doubling of computing spending in the U.S.,” Mr. Wang stated.
Japanese electronics engineer Li Jixin told Epoch Times, “No matter how powerful AI is, it requires extensive training and support from large amounts of data. Citizens of rule-bound countries often focus on protecting their privacy information, making it difficult for big tech companies to access large amounts of personal information for AI training. In China, citizens cannot oversee the government, so the government can freely use private information from citizens for large data training to gain an advantage in the AI field. For example, China’s (CCP’s) surveillance technology and network information censorship technology far exceed those of other countries.”
Mr. Wang suggested that to ensure U.S. leadership over the CCP in this regard, one should observe the scaling rules of the Chinese model closely.
“We need a competitive intelligence team to closely monitor various model scales and the corresponding computational requirements to understand these scaling rules,” he recommended.
Stanford University’s “2023 AI Index Report” stated, “Artificial intelligence is expected to become an essential component of economic and military power in the near future.” The development of AI will reshape geopolitical dynamics.
To curb the development of AI technology by the CCP, the U.S. has imposed export controls on high-end chips, limiting China’s access to advanced computing chips, the development and maintenance of supercomputers, and the capability to manufacture advanced semiconductors for military and other purposes. In retaliation, China restricts the export of rare minerals.
According to Mr. Wang, there are three key indicators to assess the AI ecosystems of China and the U.S.
The first indicator compares the quality of Chinese chips to American chips. The latest statistics show that in terms of computing power, the yield rate of Huawei’s Ascend 910B, produced by Chinese chip company SMIC, is only around 80% of NVIDIA chips. However, in terms of cost-effectiveness per dollar, the price of Chinese chips is actually two to three times higher than NVIDIA chips.
The second indicator is production levels. Huawei produces 100,000 chips per quarter, while NVIDIA produces approximately 1 million chips per quarter. Mr. Wang warned that the U.S. needs to closely monitor and track China once this ratio reaches one to three.
The third indicator is electricity. Mr. Wang stated that it is crucial to closely measure new power generation in the U.S. and China because, in the medium term, the real bottleneck for AI is electricity rather than actual chips.
In the field of autonomous driving, Scale AI’s data engine has propelled advancements in L4 autonomous driving technology, with its public sector data engine powering many key AI projects for the U.S. Department of Defense.
Simultaneously, Scale AI actively collaborates with the U.S. public sector. In August 2023, the company signed a contract with the Chief Digital and AI Office of the U.S. Department of Defense aimed at enhancing AI capabilities across the entire military, including the Army, Marine Corps, Navy, Air Force, Space Force, and Coast Guard.
In May 2024, Scale AI launched the AI-based decision platform Donovan, the first large language model deployed on a classified U.S. government network.