Owners and Tenants of Intelligence

The countries that own the computers will sell intelligence to the countries that do not. The gap is already measurable in power, money and use, and it is widening.

Intelligence is becoming something a country either owns or rents. The owners are few, the renters are everyone else, and the data we found says the distance between them is growing faster than most governments admit. This column sets out what is measured, what is not, and what two recent broadcasts of Politkast and Microphone and Mouse added from people who work with the technology.

THE GAP IN POWER

Bruegel, in a September 2026 policy brief, puts operational AI computing capacity at about 35 gigawatts in the United States (78 percent of the world), about 5 in China (11 percent) and about 2 in the European Union (nearly 5 percent). For 2031 it projects 21 gigawatts for the EU, which would lift its share only to 5.6 percent, while China rises to 15 percent. Five planned European gigafactories supply about 4 percent of that target.
The International Energy Agency expects data centre electricity use to go from about 415 terawatt-hours in 2024 to about 945 in 2030. The United States adds about 240, China about 175 and Europe just over 45.
The yardsticks differ. SemiAnalysis counts total data centre capacity of about 56 gigawatts in the United States and 24 in China (September 2026). ABI Research counts only AI-dedicated capacity and gives the United States 8.2 gigawatts in 2026. All sources put the United States first, China second and Europe far behind. They disagree on the multiple, so we quote each with its definition and do not mix them.

THE GAP IN MONEY

The Stanford AI Index 2026 reports private AI investment of 285.9 billion dollars in the United States in 2025 against 12.4 billion in China. It warns that the Chinese figure understates state-guided funds, which it estimates at 184 billion dollars between 2000 and 2023.
The best American and Chinese models are now 2.7 percent apart in performance. The distance that matters is not between those two, but between them and everyone else.

THE GAP IN USE

Microsoft's telemetry shows AI used by 28.8 percent of working-age people in the Global North and 16.2 percent in the South in June 2026, a gap of about 12.6 points. In the first quarter the North was growing more than twice as fast. Microsoft cautions that the figures describe early adopters.
Africa holds about 0.6 percent of global data centre capacity with nearly 18 percent of the world's population (World Bank). Africa and Latin America together hold 3.3 percent (a UN official at the UNCTAD investment forum). About 5 percent of African data centres are ready for AI.

WHO IS NOT AT THE TABLE

UNCTAD's 2025 Technology and Innovation Report counts 118 countries, mostly in the South, absent from international AI governance talks, and says fewer than a third of developing countries have a national AI strategy. It estimates that up to 40 percent of jobs worldwide could be affected.
The IMF found in 2024 that exposure of jobs to AI is highest in rich economies (about 60 percent), lower in emerging ones (about 40) and lowest in low-income countries (about 26). Poor countries are less exposed because they are less able to use it, not because they are safe.

WHAT THE EXPERTS SAY

Danail Brezov, a mathematician and university lecturer, told Politkast that Bulgarians are quick to adopt and improve free AI tools, and that AI is already part of daily work for him and his colleagues in Europe and Israel. The weakness, he said, is capacity: Bulgaria cannot compete with the large players, small motivated teams are what works, and energy is expensive. He saw Bulgaria's place in the production chain as software and perhaps data centres run for others.

Kaloyan Tsvetkov, on Microphone and Mouse episode 106, offered the optimist's case. Some countries will specialise in IT services around hardware, as car makers consolidated into a handful over time, and that requires customers, which means renters. Tsvetkov also warned that countries that invest too early, before demand arrives, can burn money, and that markets already corrected after the American election.

WHAT IS NOT PROVEN

That the gap makes poor countries poorer. We found no measured loss of income. The data show a gap in capacity, investment and use, not yet in living standards.
That outsourcing income in countries such as India, the Philippines and Bulgaria is shrinking because of AI. We found no figures. This is a risk, not a result.
That Europe's rules slow its AI because of foreign payments. The Politkast hosts raised the hypothesis. We found documented cases of European politicians working for Russian energy companies, and no evidence connecting that to AI regulation.

A country that does not own intelligence will pay rent for it.

What follows is the editorial view.

THE VIEW FROM THIS DESK

The numbers do not say that poor countries must become poorer. They say something more precise: those that own the computers set the price, the terms and the rules, and everyone else pays. Renting is not a disaster. A Bulgarian engineer who uses a free model gets a tool that would have cost a team a decade ago, and the optimists are right that nobody bakes all their own bread.

But rent flows one way, and the owner decides what the tenant may do. Korea, Taiwan and China did not become rich by buying technology. They learned to make it. The question for the countries that have not joined is whether they will do that, and the clock is short: Europe plans to reach 21 gigawatts by 2031, and its own numbers say its share will barely move.

For Bulgaria the choice is plain. It has cheap labour with good mathematics, and expensive power. It can be the place where small teams build on someone else's foundation, and that is honest and profitable work. It should not mistake it for ownership.

WHAT TO WATCH

Whether the EU's 2031 target of 21 gigawatts is funded and built, or remains a statement of intent.
The next Microsoft diffusion report: does the gap between North and South keep widening?
Whether the cost of using AI keeps falling, which would soften the dependence, or begins to rise.
Which countries outside the United States and China build capacity of their own, and with whose chips.
Evidence of income lost in outsourcing countries, the measure that would turn a risk into a result.