Artificial intelligence runs fast, but leaves half the world behind

Microsoft's new report photographs a two-speed technological revolution: 1.2 billion users in three years, but four billion people still excluded. Italy at 25.8%, above the global average but distant from European leaders

by Marco Trabucchi

21 November 2025

3' min read

Translated by AI

Versione italiana

3' min read

Translated by AI

Versione italiana

Artificial intelligence is officially the fastest growing technology in human history: in less than three years it has reached 1.2 billion users. This is an unprecedented rate of adoption: faster than the smartphone, the home computer and the internet itself. This is certified by the first AI Diffusion Report published by the Microsoft AI Economy Institute, which analyses AI penetration in over 100 countries across three dimensions: who develops the most advanced models (Frontier builders), who owns the computational infrastructure (Infrastructure builders) and where AI is actually used (Users).

The picture that emerges is of a two-speed technological revolution. In the global North, AI adoption is about twice as high as in the global South. WhileArab Emirates (59.4%), Singapore (58.6%) and Norway (45.3%) lead the world rankings, half of the global population - four billion people - still lacks the basic requirements to access artificial intelligence: reliable electricity, internet connection, basic digital skills.

Adoption of AI and Italy's position

AI is not a novelty relegated to techies and start-ups: it has already entered the daily routines of one in four workers in many advanced countries. With an adoption rate of 25.8% of the working-age population, Italy is slightly above the average of the global North (23%) and in line with the United States (26.3%) and Germany (26.5%). However, the gap with its European neighbours remains significant: France (40.9%), Spain (39.7%) and the United Kingdom (36.4%) show much higher penetration rates, driven by public investment, solid infrastructure and already high digitisation rates.

The report highlights how successful adoption does not necessarily require the development of proprietary models. Only seven countries in the world - The US, China, France, South Korea, the UK, Canada and Israel - host organisations that create frontier AI models. The example of Singapore, which achieves adoption rates of 58.6% without developing its own models, shows how targeted investment in infrastructure, training and coordinated policies can accelerate deployment even in the absence of advanced research capabilities.

The concentration of computational power

The infrastructure needed for AI remains highly centralised. The US and China together control 86% of global data centre capacity, with the US leading with 53.7 gigawatts, followed by China with 31.9 GW. Europe as a whole barely reaches 11.9 GW, with the EU accounting for only a fraction of global computing capacity.

This concentration creates significant barriers for countries that want to participate in the AI revolution.

The report identifies five essential 'building blocks': 1) electricity; 2) data centre; 3) internet connectivity; 4) digital skills; 5) language. The latter is particularly critical: while English dominates training datasets with over 50 per cent of web content, the 7,000 languages spoken around the world are largely absent from AI models, limiting access for billions of people.

Lessons from history

Microsoft cites the historical example of South Korea and the Philippines to illustrate the importance of technology adoption. In 1960, both countries had a GDP per capita of about $2,000. But while Korea strategically embraced industrialisation and semiconductor technology through public-private partnerships, the Philippines lagged behind. Today, the gap is abysmal: South Korea is a global technology powerhouse.

For Italy, the theme is twofold. On the one hand, the country shows a good propensity for adoption: tertiary and service sector companies are experimenting with productivity-generating tools, while universities and research centres are accelerating training. On the other, two critical issues emerge: the lack of digital skills and the slow expansion of the national cloud infrastructure.

Without these elements, AI risks remaining a predominantly individual use, not a factor in systemic transformation. As the report points out, 'the gap being created today will define who will benefit from AI for decades to come'.