China wants to build a powerful AI chip to challenge NVIDIA status

According to foreign media reports, the Chinese government introduced a new strategy in July this year. Its goal is clear: to catch up with the United States in the development of artificial intelligence technology within three years, and become the world leader by 2030. Last month, the Ministry of Science and Technology identified some details of the government plan. In this way, as a leading supplier of machine learning chips, Silicon Valley's Invitro has reached the goal of China's advancement in the field of artificial intelligence.

The document issued by the Ministry of Science and Technology lists 13 “transformation” technology projects, and plans to invest more government funds in the next few months to complete the project handover by 2021. One of them is the development of a new type of chip that runs artificial intelligence neural networks. Its software development coincides with the planning of artificial intelligence in large technology companies such as Google.

On the hardware side, a standard for the project is specifically for NVIDIA: the Ministry of Science and Technology said that in terms of chip performance and energy efficiency, it hopes to develop chips that are 20 times more powerful than NVIDIA M40 chips. These chips are called neural networks. accelerator". The M40 chip has been on the market for two years and is not the latest and most powerful chip from NVIDIA, but it is still used in various artificial intelligence projects.

The Chinese government has already targeted the NVIDIA. The research proposal put forward by the National Development and Reform Commission in October also includes the development of another high-efficiency artificial intelligence chip. In August, under the leadership of China Development & Investment Corp.'s investment fund, Beijing-based artificial intelligence chip startup Cambricon received $100 million in financing. Earlier this month, "Cambrian" announced the launch of two server chips, or will replace the NVIDIA chips used in some artificial intelligence projects.

"Cambrian" is just one of the largest Chinese companies and startups that test the artificial intelligence chip business. In the United States, there are also many artificial intelligence startups or large companies such as Google that are challenging Nvidia. In October, Horizon RoboTIcs, founded by Baidu, raised $100 million, and Deephi's deep learning platform received $40 million in financing. In addition, Huawei is working with the "Cambrian" to develop artificial intelligence chips for mobile phones and other devices.

Chinese officials and technology companies have plenty of reasons to go against the standard. To date, NVIDIA has provided hardware for artificial intelligence projects and has built a large and lucrative city. As more and more companies start investing in artificial intelligence, NVIDIA's stock market value has increased by 10 over the past three years. Times. NVIDIA supplies chips for robots, drones and self-driving cars, and has signed partners such as Toyota and Volvo.

Elsa Kania, vice chairman of the US Security Center think tank, said that from a government perspective, China has to support a strong domestic supplier, in part because of the security aspects of using foreign chips for military and other applications. Worry.

Of course, China still faces many challenges in realizing the dream of artificial intelligence software and hardware. Over the years, more computer science and machine learning research papers have been published in China than in the United States. But Caña pointed out that China's expertise in advanced artificial intelligence projects is still inferior to the United States.

Over the years, China has been striving to catch up with the US, South Korea and Japan in the chip industry. In order to independently manufacture replacements for Intel and other US processor chips, China has developed world-class supercomputer chips, but it has not been widely used on servers and personal computers. For example, as China's largest cloud service provider, Alibaba is still relying on Intel and NVIDIA chips. Paul Triolo, who tracks China's technology and related policies, said: "The Chinese are very ambitious, but the Chinese are still behind in designing and developing chips, building fabs and mass production of chips. generation."

China’s move in the chip industry has made the US government more vigilant, and there are often many obstacles to the review of US semiconductor technology. US semiconductor technology is also beginning to become more strategic. In December 2016, Obama prevented Chinese funds from acquiring US chip companies; President Trump revoked a similar acquisition agreement in September. This month, a legislative committee of bipartisan legislators introduced bills to strengthen the committee's authority to advise on similar decisions, in part because of China's ambitions in the chip industry and artificial intelligence.

But for now, despite the obvious interest of the government, China's artificial intelligence chip companies have no intention of competing for the position of NVIDIA's core chip provider. Start-up companies, Horizon Robotics, Deepsight Technologies, and Huawei, all focus on chip applications—incorporating computer vision technology into devices such as cars and cameras.

Chris Rowen, an investor in Silicon Valley chip development, said the development of the security surveillance market is also a driver of this trend. Putting an artificial intelligence chip into the camera allows it to automatically recognize people, objects or movements. Google’s recently touted Clips camera has image recognition.

Luo Wen said that China's chip startups are also focusing on the application of artificial intelligence to household appliances, auto parts and other equipment. He pointed out that "the best way to make this technology popular is to reduce costs."

Although China and the United States have already begun to compete in the field of artificial intelligence, there is no clear line between the two companies. The state-backed "Cambrian" company is getting design licenses from Silicon Valley chip developer Arteris to move the data on the chip to the "backbone" of the interconnect. Intel also invested $100 million in Horizon Robotics. The company is currently developing artificial intelligence chips for autonomous vehicles, and Google has its own products in this area. In September, Nvidia CEO Jensen Huang announced a new deal with Chinese Internet giants Alibaba, Baidu and Tencent. Trillio said: "It is an indisputable fact that China's artificial intelligence applications will run on US hardware for a while."