The AI divide.

 

History shows that technological shifts generally begin with upgrades in hardware and infrastructure, for example, from mainframes to personal computers, from landlines to mobile devices and from intranets to the Internet. This enables additional capabilities, including software and services, and facilitates the adoption and further development of technologies. The different phases are not linear; they usually overlap and create feedback loops that take years to mature and for society to realize their full potential. Currently, the diffusion of AI applications is associated with investment to upgrade critical AI infrastructure components such as semiconductors, data centres and supercomputers. These support high-speed processing, significant data-handling and advanced computation. During a gold rush, the most likely winners are often those who sell shovels. In the AI boom, one of the main winners has been Nvidia, the world’s largest semiconductor company. In 2023, based on high expectations of revenue growth, its market capitalization more than tripled to $1.2 trillion, and it nearly tripled again in 2024. The surge in AI has also benefited other top semiconductor companies, which have experienced significant growth since 2023, notably, Advanced Micro Devices, ASML, Broadcom, Samsung and TSMC.

Most of the leading semiconductor companies are from the United States and other developed economies, and there is a remarkable divide between developing and developed countries in other components of AI infrastructure. The United States has around one third of the top 500 supercomputers and more than half of overall computational performance. China ranks second, with 80 of the top 500 supercomputers, although its total computational performance is less than one tenth that of the United States. A similar situation is seen with regard to data centres, with most of them located in the United States. Few developing countries have powerful supercomputers or large data centres, apart from Brazil, China, India and the Russian Federation. Most developing countries have limited capacities in AI hardware and infrastructure, which hinder their adoption and development of AI. 

The market of AI services providers is also dominated by companies based in the United States, for example, Amazon, Alphabet, IBM, Microsoft and OpenAI, and by those based in China, including Baidu and Tencent. The private sector is responsible for most frontier AI research and produces most machine-learning models, leaving Governments and academia some way behind, with less than half combined. This is partly because of escalating costs. Since 2016, the cost of training frontier AI models has increased 2.4 times per year. More than half of the development cost is directed to hardware, making frontier AI model training unaffordable for all but the most well-funded organizations. Most SMEs, particularly those in developing countries, are unlikely to develop new AI models from scratch. Instead, they can adopt and adapt existing AI technologies to meet their particular business needs. Through interactions with numerous users and devices, companies are building up valuable data sets, enabling them to extend their advantages from hardware to data and beyond. This concentration of computing power and services in a few countries has raised concerns about their impacts on the national interests of other countries, particularly because of supply chain vulnerabilities and the interest of Governments to achieve autonomy in the development of technologies that are crucial for advancing national developmental goals.  

The United States leads the world in terms of private investment in AI, at $67 billion in 2023, or 70 per cent of global AI private investment. The only developing countries with significant investments were China in the second position, with $7.8 billion, and India in the tenth position, with $1.4 billion. In 2023, the United States also continued to lead in terms of the total number of newly funded AI companies, around seven times the number in the next highest country, China.  

Startups are key drivers of technological developments and the most valuable AI startups are primarily located in the United States and China.

Over the period 2000–2023, China and the United States were responsible for around one third of global publications in AI and 60 per cent of patents (figure I.14). Apart from China and India, most developing countries have had limited progress, and the distance from developed countries has increased. The situation is similar with regard to GenAI, with most such technologies invented in China and the United States. There is a corresponding gap in AI talent distribution; around half of the world’s top-tier researchers in AI originate from China, followed by 18 per cent from the United States and 12 per cent from Europe. The AI-related breakthroughs in recent years could mark the beginning of a new industrial revolution. AI has emerged as a general-purpose technology that can revolutionize processes in various areas powered by highly connected and intelligent production systems that can augment rather than replace humans through improved human–machine interaction. In principle, the use of AI could also help accelerate progress towards the achievement of the Sustainable Development Goals.

 Yet there are risks and ethical concerns arising from the use of biased training data and the invasion of privacy, as well as security threats, cyberattacks or autonomous weapons. If AI is unevenly distributed and lacks ethical oversight and transparency, its use may exacerbate existing inequalities, hindering sustainable human development. In addition, with high computational demands, AI consumes significant amounts of electricity and Amounts of water, with significant implications for climate change. This highlights the need for environmentally sustainable and inclusive digitalization strategies. Developing countries urgently need to strategically position themselves to harness the benefits of the AI era, while addressing potential risks and promoting equitable and inclusive AI development.