‘This Whole Thing’s Crumbling’: Scott Galloway Says AI Capex Now Rests on Just 2 Companies and Feels Exactly Like 1999

Building an AI empire takes an extraordinary amount of computing power. Finding enough customers to pay for it might be the harder part.

New York University professor and "Prof G Markets" podcast co-host Scott Galloway sees a familiar problem emerging in the AI industry. The trillions of dollars being poured into infrastructure increasingly depend on demand from just two companies, OpenAI and Anthropic.

"This sent chills down my spine. I had real déjà vu of ’99 again because this whole thing’s crumbling," Galloway said in a July episode of ‘Prof G Markets.’ "All of the demand that trickles down to the trillions of dollars of CapEx and infrastructure is now basically down to demand being created by two players: Anthropic and OpenAI."

His comparison with 1999 comes as technology companies continue spending heavily on data centers, advanced chips and the infrastructure needed to power artificial intelligence. But Galloway’s concern is that the customer base supporting those investments may be much narrower than the spending suggests.

Meta’s Infrastructure Pivot Raises Questions

Galloway pointed to Meta Platforms Inc. (NASDAQ:META) as an example of how the industry’s demand expectations may be changing.

The company has been exploring ways to lease computing capacity to outside customers, creating another potential revenue stream from its massive AI infrastructure investments.

"Meta have basically said: ‘Oh, we overestimated the front-end demand we could create so we’re going to lease out our infrastructure.’ They’re basically getting out early of these capex wars," Galloway said.

The shift raises questions about whether Meta can generate enough internal demand to justify all the computing capacity it has been building.

Leasing that capacity doesn’t necessarily mean Meta is retreating from AI. Outside customers could help monetize infrastructure that would otherwise sit underused. But the move also highlights the importance of finding customers willing to pay for the enormous computing resources being developed.

The 1999 Comparison Comes With a Familiar Risk

During the dot-com boom, telecommunications companies spent heavily building internet infrastructure in anticipation of explosive demand. When customer growth and revenue failed to meet expectations, excess capacity and financial losses followed.

Galloway sees a potential parallel in today’s AI spending cycle. Companies throughout the supply chain are investing against expectations that demand for computing power will continue growing at an extraordinary pace.

If OpenAI or Anthropic reduce their spending, the consequences could extend well beyond the two companies. Chipmakers, data center operators, construction firms and lenders could all face pressure if infrastructure investments outpace actual demand.

The AI industry could still deliver enormous long-term growth. But Galloway’s warning raises an uncomfortable question for investors: What happens when trillions of dollars in infrastructure investment depend on just two customers?

Photo via Shutterstock