UN Warns Africa and Latin America Hold Only 3.3% of Global Data Center Capacity; Closing Gap Requires $2.8 Trillion

On October 25, 2026, at the UNCTAD World Investment Forum in Doha, Qatar, International Telecommunication Union (ITU) Deputy Secretary-General Tomas Lamanauskas issued what is so far the most quantified warning on AI inequality: Africa and Latin America together hold only 3.3% of global data center capacity, 2.2 billion people worldwide still lack internet access, and closing this infrastructure gap by 2030 will require about $2.8 trillion. According to Al Jazeera, Lamanauskas said during the forum that "the truly worrying thing about the AI divide is that it is not just a digital divide—AI has enormous potential in health, education, public services, industry, agriculture and trade, and a huge divide will truly mean an opportunity divide, and will deepen socioeconomic inequality."

The Infrastructure Logic Behind 3.3%

To understand this figure, one must first understand the role of data centers in the AI era. Modern AI models—whether large language models or image generation systems—rely heavily on GPU clusters and high-speed network interconnection during training and inference. A large-scale AI training task may require thousands of GPUs working together for weeks, which means data centers must not only be large in scale but also have stable power, high-bandwidth fiber access, and sophisticated cooling systems. These conditions are relatively easy to meet in regions with well-developed infrastructure, but for most of Africa and Latin America, where power supply is unstable and network backbone coverage is insufficient, each one is a hard threshold.

The geographic concentration of computing power is not accidental. Commercial data center investment follows a typical Matthew effect: places with many users attract more investment, more investment brings lower latency and better service, which in turn attracts more users. This positive feedback loop has been running for two decades in North America, Europe, and East Asia, while Africa and Latin America were already in a catch-up position in the mobile internet era, and the gap has widened further in the AI era.

The language data Lamanauskas provided is especially specific: currently only 7% of African languages appear in AI translation tools. This is not just a matter of translation convenience; it reveals a fundamental flaw in the technology chain—the capability boundary of AI models is determined by training data, and the collection, annotation, and cleaning of training data also require computing infrastructure. There is a direct causal relationship between compute lowlands and language coverage: without local data centers and sufficient network connectivity, it is difficult to collect local language data at scale, and support for these languages in language models will remain marginal for a long time.

Who Will Be Marginalized in the AI Division of Labor

The generative AI usage data cited by Lamanauskas reveals a divergence that has already occurred: 29% of the working-age population in developed countries uses generative AI, compared with only 16% in developing countries. This gap directly affects labor productivity. When an American programmer or British analyst uses AI every day to improve work efficiency, while their African counterparts cannot benefit equally because of network speed, insufficient language support, or limited tool accessibility, the gap in individual productivity will accumulate over several years into a gap in corporate competitiveness, and then evolve into a gap in national economic output.

The impact on the job market is equally asymmetric. According to relevant data, AI is expected to displace 90 million jobs globally while creating 170 million new ones, and 40% of the skills of the working-age population will change by 2030. But the global statistic of a "net increase of 80 million jobs" obscures structural problems: the displaced jobs are heavily concentrated in manufacturing and low-skilled services, the two sectors that are the mainstay of employment in developing countries; while new jobs are highly concentrated in fields requiring computing support and digital infrastructure. Sixty percent of jobs in advanced economies face AI impact, and 40% of jobs in emerging economies face the same shock—but workers in developed countries have first-mover advantages in skills transition and access to tools, while workers in developing countries face the dual risk of "being displaced yet unable to move up."

For tech companies and the AI developer ecosystem, this data also sends a business signal: the continents home to more than 3 billion people are almost blank at the data center level. From the perspective of cloud service providers, this is both a market opportunity and a hard threshold for upfront infrastructure investment; from the perspective of AI application developers, deploying in Africa or Latin America entails inference latency several times higher than in North America and less stable service continuity. These friction factors directly suppress the formation of local AI startup ecosystems.

$2.8 Trillion: What the Number Itself Shows

The $2.8 trillion gap estimate given by Lamanauskas is the most specific official quantification to date. It should be noted that, according to reports, the detailed methodology behind this figure has not been made public, and which investment categories it covers (data center construction, submarine cables, power infrastructure, local skills training, etc.) cannot currently be fully confirmed from public materials. But the figure itself provides an order-of-magnitude reference: $2.8 trillion is several times the total annual GDP of the entire African continent. This means that relying solely on local government finances and private capital, it is almost impossible to fill this gap within four years; only large-scale investment from multilateral development banks, international aid mechanisms, and major technology companies could structurally change this landscape.

The United Nations Development Programme (UNDP) issued a similar warning in a December 2025 report, noting that countries lacking stable power, limited connectivity, and insufficient technical skills risk missing out on the economic dividends of AI. The ITU's statement this time goes further in numerical precision, upgrading the concern that "developing countries are being left behind" from a qualitative description to a quantifiable indicator that can be checked against reality.

Precedent Comparison: A Historical Mirror of the Mobile Internet Wave

The extreme concentration of the computing landscape is not unique to the AI era, but its depth of impact exceeds that of any previous technology wave. In the mobile internet era, developing countries relied on "leapfrog" paths—jumping directly from feature phones to smartphones, and from fixed broadband to 4G—to narrow the gap with developed countries to a considerable extent. The spread of mobile payments in Africa is a typical case: in regions with very low bank penetration, services such as M-Pesa achieved large-scale digital payments earlier than in Europe and the United States.

But the "leapfrog" path is harder to replicate in the AI era. The endpoint of the mobile internet was the mobile phone, whose cost kept falling and which did not depend on local infrastructure; the core production factors of AI are GPU clusters and data centers, whose capital thresholds are far higher than mobile phone base stations, and whose marginal costs do not fall rapidly over time. More importantly, gaps in AI capabilities will directly translate into gaps in the quality of public services such as language, healthcare, and education—exactly the areas where developing countries most need AI empowerment.

One specific comparison is language coverage. Major languages such as English, Chinese, and Spanish are backed by massive training data and computing power, and model capabilities have approached or even surpassed human expert levels; while only 7% of Africa's thousands of indigenous languages appear in AI translation tools, meaning that even with the same mobile devices, the quality of AI services available to African users is fundamentally different from that available to English-speaking users.

Key Signals to Watch Next

The $2.8 trillion gap figure will become an anchor for multilateral negotiations. This quantified statement means that we can expect to see countries setting agendas around "AI infrastructure assistance obligations" in multilateral settings such as the UN and G20—previous discussions remained largely within the vague framework of the "digital divide," but with a concrete figure, negotiating parties will face clearer pressure over responsibility allocation. Whether multilateral development banks such as the World Bank and the Asian Development Bank launch dedicated AI infrastructure financing windows in the next 12 months is a direct indicator.

The data center investment paths of major technology companies will face more direct political pressure. In recent years, Microsoft, Google, Amazon, and other companies have shown a diversification trend in their data center layouts in Southeast Asia and the Middle East, driven partly by sovereign data regulations and partly by user growth in emerging markets. The ITU's quantified warning provides leverage for African and Latin American governments in negotiations, but computing investment essentially follows commercial returns; if local users' willingness to pay and policy stability cannot reach the threshold, private capital will not automatically fill this gap.

The "AI sovereignty" narrative will be further strengthened at the policy level in developing countries. Lamanauskas's remarks came against the backdrop of intense bargaining among countries over AI computing power and data sovereignty, and the 3.3% figure gives the group of developing countries stronger arguments in international negotiations. Of the 104 countries that already have AI strategies, how many will include "local data center construction" as a core indicator in their national AI strategies, rather than leaving it only at the level of policy documents, is a direct indicator to follow.

Computing power has never been merely an engineering issue; it determines who can become a producer in the next round of technological dividends and who can only become a consumer. If the current 3.3% cannot be changed structurally, the AI capability gap between the North and the South will continue to widen in the coming years as model capabilities improve exponentially—and by then, the cost of catching up will only be higher than it is today.