“I do believe there’s a bubble forming with AI and data centers,” he said. “No-one’s been able to really monetize on the AI trade yet. The hyperscalers are still trying to figure it out. The semiconductors are the ones making all the money right now – they’re getting all the cashflow because the hyperscalers are buying those chips.”
While investors and financial markets are scrambling to quantify the full potential of AI, Friedman said he’s not convinced oof the actual terminal value of data centers.
Every cycle of AI development brings more efficient chip architectures, he said, potentially reducing the computing requirements that make today’s data center investments look indispensable.
“These are huge investments that are being made with the idea that this is going to be needed 20, 30, 40 years from now,” Friedman said. “No-one can give clarity if you’re going to need these data centers five to 10 years from now.”
He sees parallels between the current market euphoria and the fiber-optic boom in the late 1990s, when billions were poured into network infrastructure that became stranded far sooner than investors had expected.
