IN FOCUS: It’s the halfway mark of China’s roadmap to lead the world in AI by 2030. How is it faring?

SINGAPORE: Thirty-five seconds was roughly how long it took for a four-second video to emerge of a bustling street in Tokyo during winter, with people shopping at roadside stalls or just enjoying the snow.
The catch – everything was generated by artificial intelligence (AI), using homegrown text-to-video tool Vidu made by Chinese startup Shengshu AI.
In Wuhan, 500 AI-powered robotaxis ply the roads, putting into sharp relief a growing reality that not too long ago belonged firmly in the realm of science fiction. Similar moves are being rolled out in other Chinese cities as well.
And just last month at the World Artificial Intelligence Conference (WAIC) in Shanghai, the trailer of a new five-episode AI-generated series inspired by the Chinese literary text Classic of Mountains and Seas or “shanhaijing” was screened to industry representatives. 
Chinese strides in AI that hold their own to Western advancements – such as the likes of chatbot ChatGPT and text-to-video app Sora – are becoming increasingly clear, analysts say, as the world’s second-largest economy banks big on the technology in shaping an uncertain future.
China has set its sights on becoming the world leader in AI by 2030 and establishing itself as the premier innovation hub leading the charge in research and application. Much money and support has been channelled towards the cause.
The goal was set in 2017 when China introduced the “Next Generation Artificial Intelligence Development Plan”, which aimed for the country’s AI theory, technology, and applications to reach world-leading levels by 2030.
With the country now hovering around the halfway mark in achieving its goal, what remains for China to secure its position at the forefront of the international AI race?
While observers note the progress made, they caution of speedbumps in the road ahead – geopolitical tensions with the West being a key one.
“The primary challenge for China’s AI development at this moment lies in hardware, particularly in advanced AI chip design and manufacturing,” said Dr James Pang, an associate professor at the National University of Singapore (NUS) Business School.
“Due to geopolitical tensions, China faces restrictions in purchasing the most advanced AI chips from the US, which hampers its progress.”
At the same time, analysts believe there could be a silver lining. The curbs could provide added impetus for Beijing to produce homegrown high-tech chips and potentially supercharge its AI capabilities.
China has provided substantial government support and carved out strategic initiatives to bolster its AI capabilities, with efforts dating back years.
The now-unmentioned “Made in China 2025” policy singled out AI as a key building block in enhancing the country’s economic prowess when the 10-year blueprint was rolled out in 2015.
The plan set out in 2017 by China on becoming the global AI leader adopts a three-step roadmap. First, keep pace with AI advancements by 2020. Second, make major breakthroughs by 2025. Third, establish the country as the premier AI player by 2030.
China’s core AI industry was worth a total of 578.4 billion yuan (US$80.98 billion) at the end of last year, with an on-year growth rate of 13.9 per cent, according to multiple local reports.
Analysts have pointed out that the national AI push also ties in with a wider high-tech drive, to raise productivity and lift a stuttering Chinese economy into the next stage.
“New quality productive forces” or “xin zhi sheng chan li” in Mandarin has been a key catchphrase of the Chinese government since President Xi Jinping coined it last September. Analysts have said the term essentially refers to innovation in advanced sectors such as AI and big data.
To this end, China has recently intensified its focus on the AI-powered digital economy. A report from the China Academy of Information and Communications Technology projected that the value of China’s digital economy could reach 70.8 trillion yuan by 2025.
Agriculture, industry, and services have been the historical drivers of China’s economy. Official data shows that last year, the services sector accounted for 54.6 per cent of GDP, with industry at 38.3 per cent and agriculture at 7.1 per cent.
On Jul 29, authorities announced that the value-added output of core industries in the digital economy now accounts for 10 per cent of GDP.
Dr Li Haizhou, XQ Deng presidential chair professor from The Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen), believes China is working towards making the digital economy the top contributor to GDP within the next decade.
“So moving down this path of AI, this course is definitely the driving force, you always invest into growth … so the digital economy will be China’s largest contributor to GDP,” he said.
1979: Foundations are laid after Deng Xiaoping’s sweeping economic reforms single out science and technology as a driving force. Notable individuals like Wu Wenjin and Zhang Bo lay the groundwork for AI research and innovation.
1997: The national “863” programme to grow high-tech capabilities lists AI as a key development field for the first time, providing financial and policy support to promote AI research. 
2000s: Advancements in computing power and internet accessibility lead to significant progress in natural language processing, image recognition and data mining. Tech companies also emerge, including iFlytek, best known for its speech recognition technology.
2011: By this time, four “AI dragons” have emerged – Cloudwalk, Yitu, SenseTime and Megvii – focusing on computer vision technology.
2012: Breakthroughs in deep learning trigger a global AI boom. Chinese tech giants like Baidu, Alibaba and Tencent establish AI research institutes, recruiting top talent and promoting innovation.
2015: In its “Made in China 2025 plan”, China’s Communist Party singles out AI as a key building block in enhancing the country’s economic prowess.
2017: The State Council lays out a three-step roadmap to become a world leader in AI by 2030, with the technology being used in everything from the military to smart cities.
2022: The launch of ChatGPT by US start-up OpenAI sparks a race in China as tech firms rush to develop domestic rivals and level the playing field.
2023: The first homegrown AI chatbots like Baidu’s Ernie Bot and Moonshot AI’s Kimi go live. At the same time, a ”war of a a hundred models” emerges, referring to the intense competition that has led to at least 100 large language models, crowding the market.
2024: Chinese text-to-video AI tools surface in the wake of OpenAI’s Sora reveal. By end-July, at least three notable homegrown tools – Kuaishou’s Kling, Shengshu AI’s Vidu and Zhipu AI’s Ying – have been made available to the public.
As China charges ahead to harness AI, a growing number of homegrown tools have emerged, with capabilities that have garnered international attention.
One sector that has caught the public eye is text-to-video models, due to the potentially game-changing impact on creative industries and the possible risks of disinformation.
In February, US AI firm OpenAI – the maker of ChatGPT – unveiled Sora, its latest tool to make instant videos from written prompts. The high-quality footage and diversity of scenes generated global buzz.
Six months on and Beijing has made significant inroads into this space. At least three Chinese tech firms have put their AI video tools in the hands of global users.
Chinese start-up Shengshu AI was the latest to do so on Jul 30 for its product Vidu, following players like Zhipu AI and Kuaishou.
Unveiled in late April, Vidu was touted as the first Chinese text-to-video AI tool with capabilities on par with Sora, with Chinese reviewers referring to it as “Made in China Sora”.
Although most reviews praised its ability to generate four-second videos within 30 seconds, they noted stability issues. Distortion was also a problem, happening particularly when people-heavy footage was generated.
Meanwhile, Kling AI received more than a million requests within a month of opening the tool for domestic signups in early June, according to data released by Kuaishou.
In comparison, Sora remains unavailable to the general public, with access only granted to a select group of testers and creatives.
CNA tested Kling AI and Vidu with several prompts used and featured by Sora in its reveal video, including the one of a bustling street in Tokyo during winter.
One of the prompts was: “Extreme close up of a 24 year old woman’s eye blinking, standing in Marrakech during magic hour, cinematic film shot in 70mm, depth of field, vivid colors, cinematic”.
In about three minutes, the five-second video was ready. The results by Kling AI and comparison with Sora’s output can be seen below.
China has also been serving up answers to ChatGPT after the US-developed AI chatbot took the world by storm in late 2022. Since then, homegrown equivalents like ByteDance’s Doubao and Baidu’s Ernie Bot have emerged, to name a few.
Aside from the generative aspect, industrial applications of AI have also turned heads in China.
Driverless taxis or robotaxis have been popping up in an increasing number of Chinese cities as enterprising companies look to accelerate development and commercialise the field.
In Wuhan, around 500 Baidu-built robotaxis have been deployed, with plans to increase that to 1,000 by the end of the year, according to reports.
As AI advances at speed, analysts have pointed out the need for robust regulations to avert the risks while hopefully reaping the rewards. Action has been taken at the global, regional and national levels.
The first-ever AI Safety Summit was held in the UK last November. It saw China and the US coming together with more than 25 other countries to affirm the safe and responsible use of AI.
Within China, authorities have been trying to set out guardrails of sorts, introducing laws and guidelines targeting different aspects of AI in recent years.
In August last year, a new law regulating generative AI came into force. The first-of-its-kind legislation imposes restrictions on companies offering generative AI services to mitigate AI-related scams and safeguard user privacy.
The country has also recently implemented targeted policies addressing industry ethics and algorithm management.
Another challenge is the growing debate – both in China and beyond – between open-source and closed-source AI models, as the trajectory and characteristics of AI development could be altered depending which way the pendulum swings.
Open source is where the source code is freely accessible, allowing anyone to use, modify, and distribute it. In contrast, closed source is where access to the source code is restricted, preventing external alterations and extensions.
Both have their merits. Open-source models allow other researchers and developers to use, learn from and build on them. Meanwhile, the proprietary nature of closed-source models reduces the risks of misuse or abuse, while also allowing for more monetisation opportunities, reports have stated.
However, both also have their drawbacks. Open-source models can face resource and support constraints as they often rely on volunteers, while closed-source models offer limited control and customisation due to access curbs on the training data and internal architecture.
The current landscape in China is believed to be mixed, with some tech players embracing the open-source model while others lean towards closed source. CNA could not find any public data on the overall domestic state of play.
At the recent World AI Conference in Shanghai, Mr Robin Li, chairman and CEO of Baidu, argued that the open-source model would become obsolete. He asserted that Baidu’s Wenxin 4.0 model offered superior performance and adaptability compared to open-source models, achieving better results at lower costs.
However, this view is contested within the Chinese AI community. Mr Fu Sheng, chairman and CEO of Cheetah Mobile and chairman of Orion Starry Sky, countered that “the open-source community would ultimately prevail over closed-source models”.
He argued that open-source models were more practical, saying: “More entrepreneurs and enterprises do not need expensive closed-source 100 billion parameter models, 10 billion parameter models are enough.”
Baidu and Alibaba were among the first Chinese tech firms to add support for Meta’s open-source LLaMA family of large language models (LLMs) to their cloud computing platforms in late April. Tencent quickly followed suit.
The debate also revolves around adaptation and originality. Chinese tech unicorn 01.AI faced controversy last year for building on LLaMA models without making it clear. The firm issued an apology and said it had no intention to mask the source of the AI model.
Analysts say while leveraging existing models to create new ones is common practice, there are potential drawbacks. 
Associate Professor Wai Kin Kong from Nanyang Technological University’s (NTU) College of Computing and Data Science noted that it could be problematic if companies lack the capability to understand or modify these models.
Prof Li observed that while Chinese AI players are engaged in both leveraging open-source models and developing their own from scratch, the local community often lacks the vision to create models where the community comes together to tackle complex programming challenges.
A proponent of the open-source movement, he emphasised the importance of adhering to open-source principles and contributing back to the community.
“We don’t need 100 large models; a few well-contributed and shared models would be ideal,” Prof Li said.
Analysts also point out that progress on the AI front might spell trouble for China’s climate goals due to the technology’s vast appetite for power.
There are use cases for AI in the fight against climate change. According to the World Economic Forum (WEF), existing AI systems include tools that predict weather, track icebergs and identify pollution.
AI can also be used to improve agriculture and reduce its environmental impact, the WEF added. This could go some way in cutting carbon emissions in China, which has a sizeable agriculture sector.
According to a Google report, AI has the potential to reduce global greenhouse gas emissions by 5 per cent to 10 per cent by 2030.
But the fact remains that AI is an energy hog, particularly its generative aspect. As demand grows, so too will electricity consumption as more data centres are set up to power the charge, reports have suggested.
The International Energy Agency forecast in January that global data centre electricity demand will more than double from 2022 to 2026, with AI playing a major role in the hike.
China is aiming to boost its aggregate computing power by more than 30 per cent by 2025, according to a multi-agency plan released last October. According to the plan, more data centres will be established across the country to facilitate businesses’ access to computing power.
Higher computing power requirements lead to increased electricity consumption and carbon emissions, highlighted Dr Zhang Yaqin, Dean of the Institute for AI Industry Research at Tsinghua University during a sub-event at the Boao Forum for Asia Annual Conference in late March.
Against this backdrop, analysts say further innovations are needed.
“The challenge now is to find a clever way to deliver the same quality of intelligence, but with lower computational cost,” said Dr Li from CUHK-Shenzhen as he noted China’s advancements in AI.
“So this is kind of a success, but also, I believe it is the same challenge that other countries are facing.”
Even as China pulls out all the stops for AI dominance, analysts warn that several speedbumps lie down the road, with geopolitical tensions – particularly with the US – presenting an outsized challenge.
Graphics processing units (GPUs) are critical in powering the cause, particularly the cutting-edge ones.
Aiming to impede supercomputing and AI breakthroughs that could benefit the Chinese military, the US has imposed export controls on advanced chips and chipmaking equipment for China over the past two years.
American tech giant Nvidia controls the lion’s share of the global chip market. Before the bans, Nvidia commanded a 90 per cent cut of China’s AI chip market, Reuters reported in January.
Analysts CNA spoke to agree that the curbs present a major challenge to Beijing’s AI ambitions.
NTU’s Dr Kong highlighted that China has also been facing challenges in developing its own advanced chips. Coupled together, these slow down their AI development, he pointed out.
While acknowledging the impact of the export controls, Dr Pang from NUS noted that Chinese companies have made considerable strides, exemplified by the launch of one of Huawei’s latest AI chips.
The Ascend 910B is widely considered the most competitive non-Nvidia AI chip available in China. It has shown performance close to Nvidia’s A100 in some tests, with some instances where it outperforms the A100 by 20 per cent, according to Mr Wang Tao, COO of Jiangsu Kunpeng Ecosystem Innovation Centre.
Prof Li from CUHK-Shenzhen believes the current setbacks are a “very temporary issue”. He told CNA many China scholars believe the country’s AI drive could potentially be supercharged instead, as domestic players work to produce their own advanced chips and cut reliance on outside sources.
A related case would be the recent OpenAI access saga. Access to the firm’s AI services, including ChatGPT and text-to-image model DALL-E, are officially inaccessible in mainland China due to the Great Firewall and non-compliance with local censorship laws.
Still, some companies and engineers have circumvented the curbs by using VPNs, also known as virtual private networks, allowing them to utilise OpenAI’s tools to tweak or get their AI apps up and running.
 
On Jul 9, OpenAI tightened access by blocking users in China from using its tools and services. The company did not elaborate on the reasons for the sudden move, which took place amid frayed China-US relations.
This led to domestic AI players banding together to help one another.
At WAIC, several Chinese AI firms unveiled strategies to support developers and businesses in transitioning to local alternatives. SenseTime, for instance, highlighted its upgraded model SenseNova 5.5, and introduced a zero-cost consultation service to help users migrate from OpenAI’s models to its own.
Alibaba reported a significant increase in the download count of its open-source model “Qwen”, which surpassed 20 million in early July, a threefold increase compared to April.  
Meanwhile, Zhipu AI, a leading domestic AI player, quickly responded with a “special house-moving plan” to facilitate the transition for OpenAI users to domestic LLMs.
Mr Zhou Hongyi, founder and CEO of cybersecurity firm Qihoo 360, stated in a Weibo post on Jun 26 that “OpenAI halting China market access will only accelerate the Chinese large language model industry’s growth”.
AI is a “winner-take-all” industry, according to Mr Shaolin Zheng, the 39-year-old Chinese co-founder and CTO of NextBillion.ai, a Singapore-headquartered AI location technology solutions company.
“For example, one model has an accuracy of 96 per cent, another 98 per cent. Although just a two percentage point difference, it has intrinsic differences. For the first, I’d have to face an error rate of 4 per cent. The other is at 2 per cent – this means my workload is halved,” he told CNA.
Mr Zheng’s company leverages OpenAI’s GPT for development, with half of its revenue coming from North America. It currently has no business in China due to strict transport regulations, although some clients are Chinese companies with an overseas presence.
“As long as it is within my budget, I will choose the best model out there, I do not care about the rest,” said Mr Zheng.
To be a world-leading AI power, China will have to overtake or at least level up with the US. 
According to multiple industry indicators and analysts, both Beijing and Washington are already on the podium in the global AI competition, although the former trails the latter in some respects.
One of them is private investments in AI. The latest AI Index report by Stanford University shows that Chinese AI startups raised US$104 billion between 2013 and 2023.
In comparison, American enterprises invested US$335 billion into AI across that same timeframe, more than three times the Chinese sum.
The US also leads the way in the number of AI startups, logging 5,509 of them over the past 10 years, according to the report. China came in a distant second with 1,446.
Similarly, Washington takes the crown in being the leading source of top AI models. Last year, 61 of them originated from US-based institutions, far outpacing the European Union’s 21 and China’s 15, stated the report.
Still, Europe and parts of Asia are also advancing rapidly in the race as the groundbreaking potential of AI becomes increasingly clear.
The United Kingdom is immediately behind China in the report, yielding 727 startups from 2013 to 2023 while logging US$22 billion in private investment.
Singapore is the only Southeast Asian country in the top 10 list. According to the report, the country allocates the highest percentage of its gross domestic product (GDP) to AI, investing $15.01 for every $1,000 of GDP.
Dr Pang pointed out that when it comes to overall AI capabilities, the US leads the way “across various dimensions, including research, application, hardware, software and talent”.
“Overall, China is behind the US by approximately one to three years,” he said.
However, a broad overview doesn’t offer the full picture, especially considering how AI touches nearly all aspects of society, noted NTU’s Dr Kong.
“AI can be applied to various areas, such as drug discovery, self-driving cars, robotics, security, green energy management, and the military,” he said.
“Many companies also use AI to increase productivity. The top AI models are only part of the (overall) market.”
Analysts have highlighted how the significant investments made by China into AI are also drawing in research.
“If you have the country invest more money, then of course, there is more research,” said Dr Li.
A global AI tracker by MacroPolo, the in-house think tank of the Chicago-based Paulson Institute, found that the US remains the leader in attracting and retaining top-tier AI expertise, with 75 per cent of elite researchers working in American institutions as of 2022. 
But the same report found that almost half (47 per cent) of the world’s top AI researchers in 2022 were from China.
China also held nine places in the top 10 institutions globally by the number of AI publications in all fields in 2021, with MIT trailing in the volume of papers.
The number of high-ranking AI institutions in China is also a reflection of the “huge investments” from both the public and private sectors in AI that “produced many (academic) papers”, said Dr Li.
“If this momentum continues, I believe that in the years to come, maybe a decade or less, China will not only continue to do well in AI implementation, continue to have this leading edge, but also in AI science.”