The New Digital Divide: How AI’s Foundations Are Leaving Developing Nations Behind

AI is rapidly transforming global development, but deep divides in connectivity, compute, data, and skills mean that most low- and middle-income countries cannot yet fully benefit. The report warns that urgent investment in these “4Cs” is essential to ensure inclusive, equitable participation in the emerging AI-driven economy.

Artificial intelligence is evolving at a breathtaking pace, and global research institutions, from the Tony Blair Institute for Global Change and the Edge AI + Vision Alliance to OECD.AI and leading academic labs, are documenting how this technological shift is reshaping societies. The Digital Progress and Trends Report 2025: Strengthening AI Foundations situates AI within a fractured global landscape, where prosperity increasingly hinges on a country’s strength in four foundational pillars: connectivity, compute, context, and competency. Whether nations capitalize on AI or fall behind will depend on their ability to invest strategically in these “4Cs” that underpin modern digital economies.

Connectivity Expands, but Unevenly

Despite measurable progress, the basic conditions required to participate in the digital economy remain strikingly unequal. By 2024, high-income countries reached 93 percent internet usage, while lower-middle-income countries reached just 54 percent, and low-income nations stagnated at 27 percent. The contrast grows sharper when considering data consumption: people in wealthy nations consumed 1,400 GB of data per capita in 2023, vastly outpacing the 400 GB in UMICs, 100 GB in LMICs, and only 5 GB in LICs. One of the decade’s most dramatic transformations is unfolding above the Earth: the number of commercial communication satellites has increased fourteenfold since 2015, enabling new forms of rural connectivity and emergency response. Countries are scrambling to modernize licensing, regulate satellite–network integration, and manage spectrum congestion. Meanwhile, global ICT goods exports fell sharply in 2023, driven by supply chain readjustments, semiconductor price drops, and geopolitical frictions, hitting major exporters like China, Korea, Hong Kong SAR, and Mexico particularly hard.

Compute Becomes the “New Electricity”

If connectivity is the on-ramp, compute is the engine, and here the world is even more divided. High-performance computing systems, data centers, cloud platforms, and AI chips are overwhelmingly concentrated in a few wealthy nations. The United States hosts 175 of the world’s top 500 supercomputers and accounts for half of all global supercomputing capacity. High-income countries collectively hold 86 percent of HPC systems and 97 percent of total compute power, while low-income countries host none. Data centers follow the same pattern: 77 percent of global co-location capacity lies in rich countries; LICs hold under 0.1 percent. The cloud market mirrors this imbalance. The U.S. supplies 87 percent of global cloud computing exports, but 84 percent of that goes to other high-income nations, leaving developing countries reliant on costly imports. Renting an NVIDIA H100-powered cloud instance can cost nearly $100 per hour, an insurmountable barrier for many researchers and startups. Environmental concerns loom large, with AI infrastructure consuming vast amounts of electricity and water, generating significant e-waste, and raising national security concerns tied to data sovereignty and chip supply chains.

A Data Landscape Dominated by English

The report emphasizes that the world’s AI systems are shaped disproportionately by English-language data. Fifty-six percent of open-source datasets on Hugging Face are in English, and nearly 98 percent of the world’s scientific literature is published in it. Although over 7,000 languages are spoken globally, English makes up 45 percent of all URLs, embedding cultural and linguistic bias deep into AI models. However, as AI shifts toward multimodal learning, video-based datasets offer a more diverse linguistic picture: only about one-fifth of YouTube content is in English, with large shares in Hindi, Spanish, Portuguese, Arabic, and Russian. The report also highlights the promise and perils of synthetic data. While it can help expand multilingual datasets, especially for low-resource languages, excessive reliance risks “model collapse,” where repeated training on synthetic outputs degrades performance. For many countries, translation remains the most practical workaround, though it introduces delays, costs, and loss of nuance.

The Human Factor: Skills, Talent, and the Global Race

The demand for AI skills is surging faster in developing nations than in wealthier ones. AI-related job postings rose 16 percent in upper-middle-income countries and 11 percent in lower-middle-income economies between 2021 and 2024, compared to just 2 percent in high-income countries. But severe brain drain threatens these gains: some nations lose three to four skilled workers abroad for every one they attract. The report argues that without deep investments in digital skills, research ecosystems, and talent retention, countries risk entrenching the inequalities of the digital age. AI is reshaping global comparative advantages, and governments that strengthen the 4Cs now will position themselves for economic transformation. Those who delay may find themselves locked out of the AI-driven future.