The rush to build AI data centers in remote locations may be misguided, according to Knight Frank’s Global Head of Data Centres Stephen Beard, who argues that next-generation AI will actually require more urban proximity than its predecessors.
“The first phase of AI, the machine learning, we as the consumers, don’t need the immediacy of the response,” Beard explains, describing why early AI facilities could be located in more remote areas. However, he warns this is changing rapidly as AI evolves toward more interactive applications.
According to Beard, the industry is entering a second phase where response time becomes critical. “As a consumer, we need the immediacy of that response,” he says, describing how newer AI applications will require the same urban proximity as traditional data centers.
Beard challenges the notion that every region needs massive AI facilities. “You’ve got to be pragmatic. The UK is a very, very small island with a very, very small population in relative terms,” he says, explaining why some announced projects may be unrealistic.
He points to a principle of relativity in data center planning: facility size should match population density. “If London had a population of 10 million and needed 100 megawatts of capacity, if Manchester had 1 million, then we need 10 megawatts of capacity,” Beard explains, illustrating why massive facilities in less populated areas may not make economic sense.
The shift toward more interactive AI is driving technical changes as well. “The only difference with AI and cloud is that AI is a higher performance type of compute which requires a different type of cooling,” Beard notes, explaining how facilities are moving from traditional air cooling to direct liquid cooling systems.
This technical evolution further reinforces the need for urban proximity, as these more sophisticated cooling systems require substantial infrastructure support.
Knight Frank is already seeing the market adjust to these realities. Beard notes that major cloud providers are increasingly focused on expanding existing urban facilities rather than building new remote locations, recognizing the need for proximity to population centers.
“Our view as a house is that actually these AI requirements need to be where the existing cloud data centers are,” Beard says, suggesting that the future of AI computing will build upon, rather than replace, existing digital infrastructure networks.
While Beard acknowledges there may be room for some regional AI facilities, he emphasizes that the industry needs to be realistic about scale and location. The future of AI computing, he argues, will likely remain concentrated in existing digital hubs rather than creating new ones in remote locations.
Explore similar articles from Our Team of Experts.
The New Orleans hospitality market is experiencing a significant shift in ownership patterns, with institutional investors increasingly dominating a landscape once controlled by local operat...
Marketing leader explains why national approaches fail in real estate’s hyper-local digital landscape . The assumption that digital house-hunting behavior is consistent across markets ...
