The Shanghai government is stepping up its efforts to attract artificial intelligence (AI) talent and investments and improve regulations with the objective of building a world-class AI hub in Pudong.
Top Shanghai officials said in the World AI Conference on Thursday that the largest commercial city in China – which hosts the conference – will gather and groom AI experts to strengthen research and development and promote the use of AI technologies in the advanced manufacturing, urban management and robotics sectors.
However, Shanghai may face a hardware problem starting later this month, as the Biden administration is set to ban exports of Nvidia’s A800 and H800 AI chips to China soon and also restrict US investment funds from putting money into China’s high-technology sector.
In recent years, China’s Huawei Technologies and Cambricon Technologies have launched their AI chips, which are said to run at speeds of 60-to-70% that of Nvidia’s A100 chip. But the manufacturing of these chips is now done by Taiwan’s TSMC and can be affected by the prospective US curbs.
Supportive measures
On May 25, the Shanghai government in a document called upon private capital to invest in new infrastructure. It said it will extend the implementation period of its policy to provide an interest subsidy of up to 1.5 percentage points to private firms that build AI facilities.
It said integrated circuit, biomedical and AI sectors are the three most important industries that will build the “Shanghai high ground.”
Currently, Chinese AI firms can still buy A800 and H800 chips, which were introduced by Nvidia last November after the company’s A100 and H100 chips were added to a US export control list last August. It is said that A800’s performance has reached 70% of the A100’s.
Media reports said last week that the US Commerce Department may also ban the exports of A800 and H800 chips to China soon. The announcement reportedly would be made after US Treasury Secretary Janet Yellen’s four-day Beijing trip ends on July 9.
Ken Hu, rotating chairman of Huawei, said on Thursday that the company has been working hard developing AI chips and industrial applications in recent years.
Hu said both Huawei’s Ascend AI chips and Kunpeng central processing units (CPUs) have already made some key breakthroughs.
According to a research report published by CITIC Securities on Tuesday, more than 20 Chinese cities have used Ascend chips in their AI facilities while Huawei now has a 79% share in China’s AI computing center market.
Some media reports said Ascend 910’s performance is about 70% of that of the A100 while Cambricon’s Siyuan 370’s performance is about 60-70% of the A100. Both chips are manufactured in Taiwan using TSMC’s advanced 7nm technology.
Made in Taiwan
In May 2019, the US Commerce Department put Huawei and its 70 affiliates on its so-called entity list, on national security grounds. Huawei then started using inventory chips plus self-developed Kirin chips to maintain its smartphone output.
In September 2020, Taiwan’s TSMC stopped producing Kirin chips. In the third quarter of last year, Huawei ran out of its high-end processors and could no longer produce 5G-capable phones.
A Hainan-based columnist published an article last month with the title “Huawei’s Ascend 910 chip is leading in AI computing. Why is it not affected by the US sanctions?”
“Why can TSMC still make Ascend 910 chips but not Kirin 5G ones for Huawei? It’s because they used different technologies,” he says in the article.
He says Ascend 910 used Huawei’s self-developed Da Vinci architecture while Kirin chips used the United Kingdom’s ARM architecture, as well as technologies of the United States’ Cadence and Synopsys.
He adds that TSMC is a Taiwanese foundry, not an American firm, so it enjoys some space to maintain a business relationship with Huawei.
But a Fujian-based writer says that, even if TSMC can continue to produce AI chips for China, Chinese firms lack the tools to develop more advanced AI chips amid the US curbs and will be left behind by Nvidia.
In May, Nvidia said it will hire 1,000 people and invest up to TWD24.3 billion (US$790 million) in its new AI research center, or AI University, in Taiwan.
World AI Conference
The WAIC has been held in Shanghai annually since 2018. This year’s event attracted a lot more attention as the AI sector has become popular after the Microsoft-backed OpenAI unveiled ChatGPT 3.5, an AI chatbot, last November. Share prices of Nvidia, which has a more than 80% share in the AI chip markets, have grown 196% so far this year.
“There is a tremendous number of very smart and talented people in China,” Tesla founder Elon Musk said in a video shown during the WAIC. “The Chinese can be great at anything if they set their minds to, not only in many sectors of the economy but also in AI. I believe that China will have strong AI capabilities.”
“We will create a good atmosphere for gathering AI talent from all over the world, and accelerate the training of a group of top-notch and urgently-needed talents,” Shanghai party secretary Chen Jining said in opening speech at the WAIC. “We will adhere to open collaboration, build a platform, encourage cooperation, and create an open-source and innovative ecology.”
“We will actively explore the application practice of general AI technologies in vertical fields such as advanced manufacturing and urban management, and accelerate the development of embodied AI and robotics industries,” Chen said. “The government will strengthen safety supervision and continue to improve the AI governance and application standards.”
Shanghai Mayor Gong Zheng said Shanghai must firmly grasp the wave of new technological revolutions such as AI to upgrade its industries and turn itself into a socialist modern international metropolis with world influence. Gong said the city aims to build a high ground for itself in the AI industry.
Read: Sanctions starting to bite Huawei 4G chips sourcing
Read: Nvidia to turn Taiwan into a world-class AI hub
Follow Jeff Pao on Twitter at @jeffpao3