The race to build artificial intelligence is also a race to secure computing power. According to a University of Oxford study, the countries and companies with specialized AI data centers now hold a major advantage, while many regions remain on the margins of global AI development.
AI compute is concentrated in a small group of countries
The Oxford researchers found that only 32 countries have specialized AI data centers. More than 150 nations have virtually no access to the "compute" required to train modern AI models.
That imbalance matters because AI development depends on more than ideas, researchers, and software. Training advanced models requires large amounts of specialized hardware, cloud capacity, and data center access. When those resources are scarce or located far away, local researchers and companies face higher barriers before they can even begin to compete.
The study was conducted by Vili Lehdonvirta, Zoe Jay Hawkins, and Boxi Wu. The team mapped AI infrastructure by examining customer websites from nine leading cloud providers. Their results show a global system led mainly by the United States, China, and the European Union.
US companies such as Amazon, Google, and Microsoft operate 87 AI hubs worldwide. Chinese providers run 39. Europe has just six. The United States is far ahead, while Africa and South America are almost entirely excluded from the data center map described by the researchers.
Why the location of AI data centers matters
AI data centers are becoming a foundation for research, business, and national strategy. If a country lacks local compute, its universities, startups, and public institutions may need to rent capacity from distant providers. That can make access more expensive, less predictable, and more dependent on decisions made elsewhere.
The source article points to examples highlighted by The New York Times. OpenAI CEO Sam Altman recently visited a $60 billion project in Texas, part of the Stargate initiative. In Argentina, computer science professor Nicolás Wolovick at the University of Córdoba conducts AI research in a converted classroom using outdated chips.
Kenyan startups face a similar constraint. Companies including Qhala and Amini lack enough local computing resources, so they work at night to rent overseas capacity when prices are lower. These examples show the practical side of the compute divide: the issue is not only who can build the largest models, but who can work consistently with the tools needed for AI research and product development.
Most of the chips used in these specialized data centers come from Nvidia. The company’s GPUs have become central to the AI boom, and the source article describes them as costly and hard to obtain. When access to those chips is limited, countries without domestic infrastructure must depend on remote data centers or alternative suppliers.
Compute is becoming geopolitical leverage
The divide in AI infrastructure is also tied to geopolitical influence. The United States and China are using their technology positions in different ways. Washington controls global access to high-performance chips through export restrictions, while Beijing offers state-backed loans to promote Chinese hardware.
In the United Arab Emirates, Chinese technology has been excluded in exchange for access to Nvidia and Microsoft products. In Africa, companies such as Huawei are trying to upgrade existing data centers with Chinese chips, according to the Times.
The source article notes that China is still behind Nvidia technologically. Even so, Lacina Koné of Smart Africa sees Chinese hardware as a practical option where other routes are limited. He says Africa is open to deals with any supplier able to deliver GPUs.
That view reflects the pressure facing countries with limited compute. The immediate need is not abstract technological leadership; it is access to working hardware. Without enough GPUs, local AI developers cannot easily train models, test systems, or build services that reflect their own markets and priorities.
Local AI needs remain difficult to meet
Several governments and companies are trying to narrow the gap. India, Brazil, and the EU are funding local data centers and AI models. In Africa, Cassava Technologies plans to build five data centers with support from Nvidia and Google.
Those efforts are significant, but they do not close the divide on their own. Cassava’s own estimate says its planned data centers will meet only a fraction of Africa’s needs. The Oxford researchers warn that without broader access to compute resources, the gap could widen further.
Co-author Lehdonvirta warns that countries with computing power could gain influence comparable to oil producers. The comparison is useful because it frames compute as a strategic resource, not just a technical input. Countries that control it can shape who builds AI, where models are trained, and which regions stay dependent on outside infrastructure.
For countries outside the current AI infrastructure map, the central challenge is access. Local data centers, AI chips, cloud capacity, and reliable financing all affect whether researchers and startups can participate in global AI development. As Koné puts it, compute is becoming the foundation of digital sovereignty.
The Oxford study’s message is clear: the AI economy is not only divided by talent or ambition. It is divided by hardware, geography, and control over the data centers that make modern AI possible.