Digital Realty CEO Says an AI Slowdown Would Not Stop Data-Center Demand

Digital Realty CEO Andrew Power is pushing back on the idea that slower frontier AI development would put the brakes on data-center construction. His argument is that the buildings underpin more than the race to train new models—and that demand for cloud services, everyday AI use and scarce capacity in major markets can keep the sector moving.

In a newly published CNBC interview, Power said warnings from Anthropic, OpenAI and xAI about slowing AI development do not mean “pencils down” for AI or the real estate supporting it. The distinction matters because AI has become a dominant source of data-center demand, even as Power says cloud computing and broader digital transformation have been overshadowed by the focus on AI.

Two clocks for the infrastructure market

Power’s position draws a line between the most advanced model-development work and the wider computing market. He said large cloud providers have had to choose between expanding commercial cloud businesses and assigning capacity to AI labs. That suggests capacity redirected away from new-model training could still find demand from other workloads—a company view, not a forecast of how much demand would remain under a material AI slowdown.

JLL’s Andrew Batson offered a related but different reason for resilience: inference. Training creates or improves a model; inference is the computing used when businesses and consumers put that model to work. Batson said the next several years of data-center growth will increasingly come from that adoption. He also said only one in four Americans uses AI daily, leaving room for use to grow even if new models arrive more slowly.

A local shortage can outlast a change in AI pace

Power also made a more grounded case: demand is not evenly distributed. In markets including Northern Virginia, Dallas, Chicago, Singapore, Tokyo, Frankfurt and Amsterdam, he said demand has exceeded supply for several years. Customers competing for space in those locations cannot necessarily shift their workloads to any other market, he said. That local constraint is central to Digital Realty’s confidence, but it does not establish that every data-center project faces the same conditions.

What supports Digital Realty’s case

• Cloud computing and non-AI digital transformation remain major drivers of Digital Realty’s business, Power said.
• Inference and broader adoption, rather than only new-model training, are expected to drive future growth, according to JLL’s Batson.
• In key Digital Realty markets, customers are competing for capacity that Power says has been undersupplied for years.

The company is backing that view with a large construction commitment. Digital Realty’s development pipeline under construction totals $20 billion, double the $10 billion it reported at the end of 2023. Power said the company has broadened private-capital fundraising, pursued one-off joint ventures and positioned its balance sheet for a possible downturn. Those steps may reduce financing pressure, but they do not settle the harder question: whether customer demand will arrive at the scale and locations assumed by today’s pipeline.

The industry forecasts cited in the interview show why that question carries weight. McKinsey estimates that total data-center demand could require nearly $7 trillion in capital outlay by 2030, with AI accounting for about 70% of global capacity demand. JLL estimates the real-estate portion alone could reach $3 trillion over the next five years. They are estimates, not committed spending—but they frame the bet that infrastructure demand can broaden faster than concern about frontier-model development narrows it.