As the AI race shows no signs of slowing down, it's clear which countries started in pole position — and are poised to hold onto their lead.
"Artificial intelligence has created a new digital divide," warns the New York Times in new reporting, "fracturing the world between nations with the computing power for building cutting-edge AI systems and those without."
That warning is based on new data from researchers at Oxford University showing the distribution of the world's most powerful data centers. It's a rare glimpse: historically, companies and nations have tried to keep information about these facilities under close wraps.
The clear leaders in the race, per the NYT, are the European Union with 28 centers, the US with 26 centers, and China with 22. Collectively, they house more than half of the world's AI-capable data centers mapped in the study. And in all, only 32 countries — and the vast majority in the northern hemisphere — boast the facilities with the computing power necessary to run generative AI systems.
"We have a computing divide at the heart of the AI revolution," Lacina Koné, the director general of the continental digital policy group Smart Africa, told the NYT. "It's not merely a hardware problem. It's the sovereignty of our digital future."
Generative AI models are ludicrously expensive to train and run. The Trump administration's exorbitant $500 billion AI infrastructure project, dubbed Stargate, is as much a testament to the ridiculous costs involved as it is to the unfettered hype for the technology.
The biggest upfront cost is the pricey graphics processing units needed to train the AI models, costing tens of thousands of dollars each. A single data center may have tens of thousands of these AI chips. The largest, like Elon Musk's "Colossus" facility in Tennessee, boasts over 200,000.
Few can afford that kind of investment. If you aren't living in one of a handful of countries where these facilities are available, then you're forced to remotely rent computing power, which is far from ideal. Qhala, a Kenya-based startup that's building a large language model focused on African languages, is facing this predicament. Per the NYT, its engineers need to cram their work early in the morning, when most American programmers are asleep, so they can experience faster speeds.
"Proximity is essential," Qhala founder Shikoh Gitau told the NYT.
"If you don't have the resources for compute to process the data and to build your AI models, then you can't go anywhere," echoed Kate Kallot, a former Nvidia executive and the founder of Amini, a competing AI startup in Kenya.
... continue reading