Self-developed Hanguang 800 is part of company’s broader push into hardware
Coco Liu, Cheng Ting-Fang, Hong Kong, Taipei, Nikkei Asian Review, September 25, 2019
Alibaba Group Holding has unveiled its first self-developed artificial intelligence chip for data centers, a move aimed at reducing the Chinese tech giant’s reliance on U.S. technology amid persistent tensions between Beijing and Washington. The announcement also marks an important milestone in Alibaba’s attempt to become a serious player in hardware. After starting out 20 years ago in e-commerce, the company is now following Google, Amazon and other Western internet rivals in attempting to move up the tech value chain with its own chip offerings. “This is just the first step of a 10,000-mile journey,” Jeff Zhang, Alibaba chief technology officer, told audiences at the company’s annual cloud computing conference in Hangzhou on Wednesday.
Credit: Fotolia
“Alibaba is confident that we are not only capable of doing what a traditional hardware company could do but also capable of accomplishing what they could not do,” Zhang said. With its hugely popular e-commerce and digital payment services, Alibaba is well-positioned to access the large amounts of data needed for training AI. Zhang said the new chip, known as the Hanguang 800, has a computing power equivalent to that of ten graphics processing units, or GPUs, a more general-purpose type of computing chip used to process AI workloads. This technological breakthrough, he added is already making a difference in Alibaba’s e-commerce business. For instance, it used to take an hour for the system powering Alibaba’s online marketplace, Taobao.com, to process a billion product images for its website. Now that it is running on the new AI chip rather than GPUs, the system can do the job in five minutes, according to the company.
“The launch of Hanguang 800 is an important step in our pursuit of next-generation technologies, boosting computing capabilities that will drive both our current and emerging businesses while improving energy-efficiency,” Zhang said. Alibaba did not disclose whether it would replace the GPUs it uses with its own AI chip on a large scale. However, the company’s billionaire co-founder Jack Ma has indicated his desire to reduce his company’s reliance on U.S. chips. Most of the processors used to compute complicated AI workloads and analyze videos, images and languages in data center servers are made by U.S. chipmakers Nvidia, Advanced Micro Devices (AMD), Intel’s Altera, as well as Xilinx. “The market of chips is controlled by America … and suddenly they stop selling,” Ma told an gathering of students at Waseda University in Tokyo last year. “Japan, China, any country should have their own technology. A company should take responsibility for its customers, for the global future,” he continued.
In April 2018, days after the U.S. banned ZTE from importing American components, Alibaba acquired Chinese chipmaker Hangzhou C-SKY Microsystems to expand its own chipmaking capabilities. Washington’s crackdown on Huawei Technologies this year has accelerated the so-called “de-Americanization” trend in China, with more and more tech companies in the country seeking to develop their own chips and other key components. Alibaba said that it has no plans to sell the chips. But analysts say equipping its cloud service with highly-efficient AI chips could give the company a leg up in attracting more customers in China’s increasingly competitive cloud market. “It’s not too surprising that Alibaba would develop its own AI chip as it deploys data centers on a massive scale across China and in some emerging markets. AI computing is quite fragmented and every tech giant has different needs,” said Sean Yang, an analyst at Shanghai-based CINNO. The pursuit of customized AI chips — designed to perform specific tasks with lightning speed and lower power consumption — rather than the general-purpose graphics processors used by most companies, has become an emerging trend among technology heavyweights worldwide.
In 2016, Google announced its first Tensor Processing Unit, or TPU, to compete with Nvidia for training AI models. The company is reportedly building a team in India to create more AI chips. Facebook, meanwhile, recently announced its AI chip ambitions, while Amazon last year unveiled its own chips as an alternative to traditional computing processors for AI projects. “The new announcement on AI chips can be viewed not only as an effort to decouple from U.S. chipmakers, but from a business point of view, it’s also a way for Alibaba to build more customized and competitive data center cloud service to compete with rivals like Amazon, Google, Tencent and Microsoft,” Yang said. Alibaba is the undisputed leader in China’s public cloud market, but its long-time rival Tencent Holdings is catching up. IDC data shows that in the segment known as “infrastructure as a service” — one of the most popular forms of cloud services in China — Tencent achieved a market share of 11.5% last year. That was far behind the 43% share of Alibaba Cloud, or AliCloud for short, but it was an marked improvement from 7.4% in 2016. Meanwhile, other Chinese technology giants are also doubling down their investment in the cloud, eyeing for a slice of this booming market. Huawei in September pledged $1.5 billion to recruit software developers for its computing systems. On the global level, Alibaba ranked the third in cloud computing after Amazon and Google.
Recent Comments