At the press conference for 2023 China Computing Power Conference, held on July 17, the director-general of China's Information and Communication Development Department, Xie Cun, observed that a gap still persists between China and its international competitors in terms of computing infrastructure.
In light of this, Xie highlighted the Department's priorities to address the gap, including strengthening R&D efforts on CPU, GPU and server products, alongside boosting innovations in computing architectures and algorithm.
Yunming Zhang, Vice Minister of China's Ministry of Industry and Information Technology (MIIT), claimed at the press conference that the number of deployed standard datacenter racks in China has already surpassed 6.5 million units up to the end of 2022, with a total computing power of 180 EFLOPS, second only to the United States. The total storage capacity is over 1,000 EB (1 trillion GB).
China Academy of Information and Communications Technology (CAICT) estimates that the scale of China's core computing power industry reached CNY1.8 trillion (US$251 billion) in 2022. Every yuan invested in computing power will drive 3 to 4 yuan in GDP growth.
Strike a balance between computing and storage
Meanwhile, at the ongoing Nanjing International Semiconductor Expo, Dr. Ni Guangnan, a leading semiconductor researcher from China Academy of Engineering, observed that the computing power of China-based data centers is relatively insufficient as they primarily use mechanical drive, while China's semiconductor storage technology lags behind. Therefore, he sees a nationwide need to replace mechanical storage with semiconductor storage, though the process is slow due to costs.
According to Ni, storage technology has become strategic as different countries step up efforts to develop their data strategies, while the data volume in China is expected to grow to 48.6ZB by 2025. Driven by generative AI, China is especially experiencing a boom in "AI computing center". However, the scholar perceives a tendency to favor computing power than storage. To be a data superpower, China has to be backed up with a powerful and advanced storage industry.
In this context, Ni calls for a development strategy that balances "computing" and "storage", suggesting a proportionate adoption of solid state drive (SSD) within storage infrastructure. In other words, government procurement and bidding process should prioritize SSDs, and the evaluation criteria should not be based solely on a single price indicator. In addition, he calls for the government to promote the use of domestically produced SSDs when it comes to important infrastructure.
Data retention and management equally important
As memory becomes a key component of the AI race between China and the US, data retention and management strategies also matter to create AI-ready data. In this sense, some believe that China - in particular its military - might have an edge over the US. Alexander Wang, founder of Scale AI, pointed out in his July 18 testimony before the US House Committee on Armed Services that China's data collection and algorithm development will likely define its advantage going forward, while the US struggles to reach equivalence in this area.
"During the daily battle rhythm, the Department of Defense creates more than 20 terabytes of data daily, and because of their outdated data retention and management policies, warfighters, analysts, and operators are unable to tap into its full potential because it is not AI-ready," Wang indicated. "The Department of Defense has been working for more than a decade to solve these complex challenges, however, more needs to be done."
