The new AI revolution is not just creating opportunities—it is quietly building a dangerous gap between countries that can shape the future and those that can only consume it. And this is the part most people miss: the real divide is no longer just about having the internet, but about who controls the foundations of AI itself.
Artificial intelligence is advancing at remarkable speed and is now central to how economies grow and societies function. Global organizations and research groups have started to map how this shift is unfolding, showing that AI success increasingly depends on four basic building blocks: connectivity, compute, context (data), and competency (skills). These “4Cs” are becoming the new markers of prosperity, and a country’s position in the AI-driven world will largely depend on how well it invests in them.
Yet most low- and middle-income countries are still far from being able to fully participate in this new AI economy. The gaps in infrastructure, data, and human capital are so wide that without urgent, targeted investment, many nations risk being locked into a permanent role as late adopters. That is why the report stresses that strengthening the 4Cs is not a luxury or a long-term aspiration—it is an immediate priority for anyone who wants AI to support inclusive and equitable development rather than deepen inequality.
Connectivity: progress with big blind spots
On the surface, the world looks more connected than ever, but the picture changes dramatically when you zoom in on who actually has meaningful access. By 2024, people living in high-income countries had reached about 93 percent internet use, while in lower-middle-income economies only a little over half the population was online, and in low-income countries, barely more than a quarter had access at all. This is not just a minor difference; it fundamentally shapes who can use AI tools, digital services, and online learning in their daily lives.
Data usage tells an even starker story. In 2023, people in wealthier countries consumed around 1,400 GB of data per person, compared with roughly 400 GB in upper-middle-income countries, 100 GB in lower-middle-income countries, and just 5 GB in low-income nations. That means an average user in a rich country may be streaming, learning, and working online at a volume that is simply unimaginable for many users in poorer regions. For AI, which thrives on data-heavy applications like video, cloud tools, and real-time services, this difference is huge.
One of the most dramatic shifts is happening in space. Since 2015, the number of commercial communication satellites has multiplied roughly fourteen times, opening up new possibilities for internet coverage in remote and rural areas and providing critical connectivity during disasters. Governments are now racing to update licensing rules, figure out how to integrate satellite networks with existing operators, and prevent radio spectrum from becoming overloaded. But here’s where it gets controversial: even with all this innovation, many communities still struggle with affordability, device access, and digital literacy, meaning satellites alone will not fix the divide.
At the same time, global exports of ICT goods fell sharply in 2023. This drop has been linked to supply chain restructuring, lower semiconductor prices, and rising geopolitical tensions, which have hit key exporters such as China, Korea, Hong Kong SAR, and Mexico especially hard. The ripple effect is that hardware needed for connectivity and AI can become more uncertain or unevenly distributed, adding another layer of vulnerability for countries already lagging behind.
Compute: the new “electricity” of AI
If connectivity is the road into the digital world, then compute power is the engine that makes AI actually run. And today, that engine is heavily concentrated in the hands of a few wealthy nations. High-performance computing systems, large-scale data centers, cloud platforms, and specialized AI chips are clustered in a small group of countries that are effectively becoming AI “powerhouses.”
The numbers illustrate the imbalance clearly. The United States alone hosts 175 of the world’s top 500 supercomputers and holds about half of the total global supercomputing capacity. High-income countries together possess roughly 86 percent of all high-performance computing systems and an astounding 97 percent of overall compute power. Low-income countries, by contrast, do not host any of these top-tier systems, leaving their researchers and innovators heavily dependent on foreign infrastructure.
The pattern is similar when looking at data centers. About 77 percent of global co-location data center capacity—facilities where organizations can rent space and services—sits in high-income countries. Low-income nations hold less than 0.1 percent of this capacity. That means most of the world’s digital processing and storage happens far away from where many people live, raising concerns around dependence, resilience, and the ability to enforce local regulations.
Cloud computing, often seen as a way to “rent” compute instead of building it locally, also reflects this imbalance. The United States supplies roughly 87 percent of global cloud computing exports, but the majority—around 84 percent—flows to other high-income countries. Developing economies therefore tend to access cloud services as expensive imports, with pricing that can be prohibitive for small labs, startups, and universities.
To put the cost into perspective, renting a powerful cloud instance equipped with an NVIDIA H100 GPU can approach $100 per hour. For well-funded labs in rich nations, this is painful but possible. For many researchers, entrepreneurs, or students in lower-income countries, it is simply out of reach, effectively excluding them from cutting-edge experimentation and innovation.
There is also a critical environmental and security dimension. Large AI data centers and training clusters consume huge amounts of electricity and water and generate substantial electronic waste. Countries must wrestle with questions such as: How can they expand compute capacity without undermining climate and sustainability goals? How do they secure data sovereignty when most compute sits abroad? And what happens if access to chips and hardware becomes entangled with geopolitical rivalries and export controls? This is where opinions may sharply diverge—some argue that aggressive AI expansion is worth the environmental trade-offs, while others see it as an irresponsible burden on already fragile ecosystems.
Data and language: when English dominates the AI world
AI models learn from data, and the kind of data they see shapes how they “understand” the world. One of the report’s most striking warnings is that today’s AI systems are built overwhelmingly on English-language material. A majority of open-source datasets on platforms like Hugging Face—around 56 percent—are in English, and close to 98 percent of scientific articles globally are also published in English. This means that knowledge, values, and perspectives expressed in other languages are heavily underrepresented in the training material that powers many AI tools.
This is especially troubling when considering that the world is linguistically rich, with more than 7,000 languages spoken across the globe. Yet English accounts for about 45 percent of all URLs. When so much of the digital universe is dominated by a single language, AI systems inevitably carry built-in cultural and linguistic bias. For people using AI in non-English contexts, the tools may feel less accurate, less relevant, or subtly misaligned with local norms and realities.
There are some promising shifts as AI becomes more multimodal and relies more on video, audio, and images rather than text alone. Video platforms, for example, show a more diverse language mix. Only about one-fifth of YouTube content is in English, with large portions produced in languages such as Hindi, Spanish, Portuguese, Arabic, and Russian. This diversity could, in principle, help train AI models that better reflect global cultures—if the data is collected and used responsibly.
The report also delves into the growing use of synthetic data—data generated by AI models themselves. On the one hand, synthetic data can help boost representation for low-resource languages by creating additional examples where real-world material is scarce. On the other hand, relying too heavily on AI-generated data can trigger “model collapse,” where systems are increasingly trained on their own outputs and gradually lose accuracy and robustness. This tension raises a provocative question: should countries with limited real-world data lean into synthetic data as a shortcut, or could that strategy backfire in the long run?
For many nations, translation remains the most practical short-term solution to access AI tools that were originally built for English. They can translate interfaces, documentation, and some content into local languages, but this method comes with clear drawbacks. Translation adds time, cost, and often strips away nuance, cultural references, and subtle meaning. It is a workaround—not a true fix—for the deeper imbalance in whose language and worldview AI actually reflects.
People power: skills, talent, and a new global race
Behind every AI system are people: engineers, researchers, policymakers, educators, and users. The report highlights that demand for AI-related skills is growing particularly fast in developing regions. Between 2021 and 2024, AI-focused job postings increased by about 16 percent in upper-middle-income countries and 11 percent in lower-middle-income economies, compared with only around 2 percent growth in high-income nations. This suggests that employers in emerging economies are eager to hire AI talent and build local capacity.
However, the talent story comes with a serious warning. Many countries face intense “brain drain,” losing more skilled workers than they attract. In some cases, for every one highly trained professional who arrives, three or four leave to work abroad. This constant outflow makes it extremely hard to build sustainable local AI ecosystems—universities, startups, research centers, and government agencies all struggle when their best people depart.
The report argues that without major, long-term investment in digital skills, research institutions, and talent retention, the digital revolution will deepen existing inequalities instead of reducing them. AI is reshaping what kinds of industries countries can specialize in, affecting everything from manufacturing and services to health, education, and agriculture. Governments that act now to strengthen connectivity, compute, context, and competency will be better positioned to harness AI for broad-based development.
Those that delay may find themselves “locked out” of the AI-driven future, dependent on imported tools designed elsewhere and forced to adapt to rules they did not help create. But here’s where it gets controversial: should developing countries focus on catching up in all four pillars at once, or should they strategically pick a few areas where they can realistically lead or specialize? Some will argue for massive, state-led AI infrastructure drives, while others will push for leaner, more targeted strategies centered on people, not hardware.
And this is the part most people miss: the choices countries make today about digital infrastructure, language, skills, and governance are not just technical decisions—they are decisions about power, culture, and who gets to define the “intelligence” in artificial intelligence. So what do you think? Should wealthy nations be required to share AI infrastructure and data more fairly, or is it up to each country to compete and catch up on its own? Do you believe the AI boom will ultimately narrow or widen the gap between rich and poor countries? Share your thoughts—do you strongly agree, strongly disagree, or see it differently altogether?
