AI in mainland China
  • Global Research
    • Disruptive technology

AI in mainland China

  • Article

The world’s faster growing artificial-intelligence market is an innovative leader too.

Artificial intelligence is an increasingly important part of daily life. AI-enabled smartphone cameras shoot better pictures. AI makes autonomous car driving a reality. In healthcare, it provides tools for diagnosis, treatment planning and rehabilitation.

City surveillance cameras monitored by humans are time consuming and prone to error: AI-enabled cameras can identify traffic conditions or fire hazards and alert humans.

The global AI software market is expected to exceed USD120bn by 2025 – a fourfold increase since 2020. But AI companies typically focus on their home markets because of concerns over data safety, patents, and regulatory restrictions. This is especially so in the world’s fastest growing major market, mainland China, which is forecast to be USD26bn by 2025.

Five prominent Chinese artificial-intelligence companies are on the US Entity List that restricts US exports to them, and one is on a US investment blacklist. But Chinese companies are leading much of the innovation in AI.

Traditionally, AI models are produced individually, which is inefficient and expensive. Also, while long-tail scenarios – such as objects falling from high-rise city buildings – are rare, they are important to factor in. However, they are uneconomic and provide little data to train the AI model.

Now though, one Chinese company’s infrastructure allows cost-efficient mass production of AI models for such scenarios using only limited data by streamlining and minimising human intervention.

AI computing chips are the foundation for training AI algorithm models. They are housed in infrastructure such as cloud-computing service providers and machine-learning frameworks help train the models.

We think application-specific AI chips will gradually replace general graphic processing unit chips, whose poorer storage expansion increases the time and cost for AI model training. And Chinese domestic chips will gradually replace foreign chips.

The image signal processing chips used to process data from sensors need to work seamlessly with the AI sensor but traditional chips accept only standard data format and cannot preserve critical information, thus limiting the output image quality. However, new Chinese versions should conduct multi-format data processing and work with new types of sensors such as depth sensors, multispectral sensors, and dynamic-vision sensors.

Car companies have been investing significantly in developing advanced driver-assistance systems. Up to 90 per cent of all vehicles sold in mainland China are expected to be equipped with Level 2 or higher autonomous-driving capabilities by 2030 – an increase over a decade from less than 9 per cent. And up to a fifth of vehicles are will have Level 4 or above capabilities.

These capabilities include adaptive cruise-control, lane-centring and traffic-jam assist, using vision-based systems that can detect vehicles at 200 metres and pedestrians at 150 metres, plus multi-sensor fusion systems such as LiDAR.

Meanwhile in-car AI can recognise drivers’ identity and detect drowsiness or distraction as well as provide information or entertainment.

But intelligent vehicles require massive computing power and storage, which is often costly and complex. These companies also need to address quickly long-tail scenarios that could be dangerous and diagnose safety problems besides updating to AI models. This is costly without an established AI platform, so Chinese software companies are developing tailored models.

First published 4th February 2022.

Would you like to find out more? Click here to read the full report (you must be a subscriber to HSBC Global Research)

Global Research

HSBC Global Research provides information, insights and thought-provoking ideas.

Need help?

For more information, please contact your HSBC representative.