Artificial Intelligence (AI) has become a strategic technological battleground between the U.S., China and other countries. The U.S. Government just pledged to increase spending next year on AI from $1.2 billion to $1.5 billion. The importance of AI to the U.S. is captured by a statement from the U.S.’ chief technology officer, Michael Kratsios, saying that the Trump administration had taken “unprecedented action to prioritize American leadership in AI and quantum.” Winning the AI race is potentially the most important challenge technologists face today. China followed suit on August 28 by adding AI to the list of technologies that are restricted or banned from export.
The United States and China find themselves in fierce competition to see which country will dominate the AI landscape and control future implementations of this emerging technology. One of the most hard fought areas in the struggle is the race to develop AI-accelerated hardware and acquire patents to protect innovations here. The current goliath from the U.S. is Nvidia, the market leader in graphic processing unit (GPU) hardware that has been used for increasing the speed of AI applications. In the past three years, however, a Chinese contender named Cambricon Technologies has become the poster child for the Chinese effort to lead in this area. From a patent innovation perspective Cambricon is dominating this battle, with more than four times as many patent inventions as Nvidia.
Chinese AI researchers and companies are traditionally thought of as innovators in the development of software. Megvii Technology and its Face++ system, for instance, uses AI and machine vision software and is recognized as one of the world’s leaders in the facial recognition field.
While software development is critical for the advancement of AI systems, specialized hardware is perhaps even more essential to ensuring that the software can provide results in a timely and comprehensive fashion. The chart below shows the progression of patent innovations published in the area of specialized AI chip development between the U.S. and China in the past ten years. China has been keeping pace with the U.S. in terms of patent innovations published per year during that time—and in 2019, it passed the U.S. for the first time.
On August 28, China revised its list of banned or restricted technologies for export from the country for the first time in 12 years and AI was on the list. Even before the restrictions, China had made the development of AI systems a national imperative and China’s State Council published a policy blueprint in the summer of 2017 that set out the goal of creating the “world’s primary AI innovation center” by 2030. The plan included ambitious goals for software development, but also covered the importance of chip development, especially chips that could be used in mobile devices. A great deal of research was carried out at the Chinese Academy of Sciences, but it’s two graduates of this university system that are making the biggest waves in the AI chip market.
Brothers Yunhi and Tianshi Chen are the brains behind Cambricon Technologies, the company providing hardware for China’s AI ambitions. Cambricon is valued at $2.5 billion and recently raised $368 million in its IPO. The company’s Cambricon-1H and Cambricon-1M chips can already be found in almost 100 million mobile devices and servers. Its flagship Cambricon-1A is used in edge devices and has been touted as the first deep learning processor for commercial use. Besides having chips that are powering servers for Alibaba and Huawei’s latest generation of AI-enabled smartphones, the company also has one of the largest patent portfolios in the field. The graph below compares the size of the Cambricon AI chip patent portfolio to Nvidia’s. The U.S. company is considered one of the leaders in the AI chip field with its Ampere AI chip.
Chinese companies are criticized for only filing patents in their home country and this leads to some speculation about how valuable these patent filings are. As seen in the graph above, Cambricon is a notable exception to these criticisms, since a substantial portion of its portfolio is made up of patents filed for in the U.S. or using the World Intellectual Property Organization international application. Even if Chinese patent filings weren’t included—and they most certainly should be—Cambricon would still have more U.S. and international filings in this area than Nvidia. The U.S. company does have a larger patent portfolio than Cambricon overall, but only a portion of it covers AI acceleration hardware.
AI hardware or chipsets are specialized microprocessors that are designed to accelerate AI software systems. For example, these customized chipsets are meant to increase the performance of artificial neural networks. Training and implementing of these systems require vast amounts of data and hardware that can process it in parallel while maintaining low power demands. Early AI chip implementations relied on general-purpose GPUs, but recent research has gone into the invention of hardware exclusively designed for running AI software.
So far, Chinese companies have been happy with feeding the growing needs of the Chinese market for AI hardware. This fits with the national imperative to become the world’s innovative leader in this technology. At some point however, as the need for AI systems grow globally, Chinese organizations may look to expand into these markets as well.
The Chinese government’s new export restrictions will make that even more challenging. Cambricon and its brethren have a home field advantage in China, but will face an uphill struggle to overcome not only internal restrictions but foreign constraints such as the ones Huawei has encountered with Western governments. Chinese companies have an advantage with the stockpile of patents they’ve generated, but it remains to be seen if they can translate this into a stunning commercial success.
No CommentsLeave a comment Cancel