Kevin O'Leary Warns AI Demand Is 'Insatiable' But 65% Of Data Center Projects Will Never Be Built

AI data centers are having a classic real estate problem. Everyone wants one, but not every plot of land can support the dream.

"Shark Tank" investor Kevin O’Leary believes 65% of announced data center projects will never be built, even as demand for computing power continues to surge.

"In my personal opinion…there’s many projects [that] have been announced and 65% of them will never be built," O’Leary said in a Marketbeat released Oct. 16. "They will never get there. They will have a fatal flaw or political reason not to be built."

He warned that 60% to 65% of projects could be shut down within the next 24 months. "And listen, I’m signaling that warning out there cuz I’m just watching it happen," he said.

Power, Water And Politics Threaten Projects

O’Leary pointed to electricity access, water, fiber connections and political opposition as reasons projects could fail. Some developers also assume they can tap natural gas infrastructure, only to find those plans won’t work.

Local resistance and state-level moratoriums can derail projects before construction begins, leaving developers with land and plans but no viable path forward.

Electricity is becoming a particularly important constraint as more data centers compete for power. The Lawrence Berkeley National Laboratory in June estimated that data centers could consume 11.8% of total U.S. electricity by 2030 Its projections ranged from 9.5% to 15.3%, depending on how demand and technology develop. 

That growing demand helps explain why access to power is central to O’Leary’s investment strategy. A site may have plenty of land, but without a reliable power source and the infrastructure to support it, construction plans can stall.

Compute Demand Could Triple By 2027

Despite his warning, O’Leary isn’t predicting a collapse in AI infrastructure demand.

"The amount of demand just for compute capacity both from the government agencies and from the private sector and the hyperscalers and others is insatiable for Q4 of 2027. Demand is probably 3x what we have right now," O’Leary said.

"So I don’t see a slowdown at all in the next 3 years at all in terms of the demand for compute power," he continued.

O’Leary is betting on the infrastructure needed to support that growth, with investments spanning the U.S., Canada, Norway and Finland. His exposure includes Bitzero Holdings (NASDAQ:AIBZ) and private development projects in Utah and Alberta, Canada through O’Leary Digital.

He favors sites with access to low-cost power, land, water rights, fiber connections and permits. A single large project can cost roughly $16 billion for about 1.4 gigawatts of capacity, he said.

The gap between announced capacity and projects likely to be completed has become a central question for the industry. 

The AI infrastructure boom may have plenty of room to grow, but the projects that ultimately deliver computing power could be a much smaller group than the announcements suggest.

Photo via Shutterstock