Watch / Artificial Intelligence (AI)
Too much AI, too little return?
Too much AI, too little return?
About this video
Key points:
Roger Montgomery warns that AI infrastructure spending is vulnerable if venture capital funding for unprofitable AI startups slows.
He argues that debt-funded hyperscaler investment and reliance on OpenAI and Anthropic create risks of a broader downturn across the AI ecosystem.
Montgomery sees the risk of excess data centre capacity and weaker-than-expected returns, comparing AI infrastructure investment with past overbuilding in the airline industry.
Roger Montgomery from Montgomery Investment Management says the AI boom rests on a potentially fragile funding structure. He argues that major hyperscalers, including Microsoft, Amazon, Google and Meta, account for a substantial share of Nvidia’s chip revenue, with infrastructure purchases increasingly funded by debt. Meanwhile, Montgomery estimates that OpenAI and Anthropic depend heavily on venture capital-backed startups for cloud computing demand. If higher interest rates make venture funding less attractive, he warns that these customers could cut spending, creating pressure across the supply chain.
Montgomery says the risk extends beyond individual AI companies to the broader investment cycle. He believes the market may be relying too heavily on the scale of infrastructure spending as evidence that the investment will ultimately pay off, despite questions about profitability. He was also sceptical of the proposed Firmus listing, saying he had advised against investing ahead of its disappointing outcome. The recent share-price decline in CoreWeave, a highly leveraged AI infrastructure provider, adds to his concerns about the sector’s financing and valuations.
If AI investment slows sharply, Montgomery expects some data centre capacity to become uneconomic or remain unused, rather than quickly turning into a valuable source of scarcity. He says any recovery through reduced supply and higher prices could take years. Drawing a parallel with airlines that overbuilt capacity ahead of demand, he argues that forecasting AI infrastructure needs is difficult: too much capacity destroys returns, while too little limits growth. Montgomery questions whether providing AI services will ultimately be as profitable as investors currently expect.