The AI Bubble Debate – Humble Student of the Markets

Are we in an artificial intelligence investment bubble? That’s becoming the narrative in the financial media. A search on Bloomberg for “AI bubble” showed elevated number of stories, with a peak in January 2025 after the news of the DeepSeek breakthrough.

In practice, what does that mean for equities, and the economy?

The Case for a Bubble

Here is the case supportive of a bubble, or at least a case of irrational exuberance.
A review of regional equity valuations by forward P/E shows that the U.S. is at the top of its 20-year range. By contrast, the forward P/E of other regions is either in the top or second quartile of their historical ranges.

The valuation gap between the U.S. and the rest of the world matters because funds flow data shows that foreigners have piled into U.S. financial assets and pushed their holdings to record levels.

Companies have caught the AI fever. Investment in information processing equipment and software represented 4% of GDP growth on an absolute basis and 92% of GDP growth in H1 2025. Excluding these categories, H1 2025 GDP grew at annualized rate of 0.1%.

Rational Exuberance

On the other hand, a case can be made that the surge in AI-related plays is a case of rational exuberance. Ed Yardeni pointed out that today’s “S&P 500 Information Technology and Communication Services sectors account for a record 44.9% of the index’s market capitalization but also a record 37.4% of the index’s forward earnings. During the Tech Bubble of 1999-2000, their combined market cap and forward earnings shares peaked at 40.7% and [only] 23.8%.”

The markets are just responding to the promise of AI as a revolutionary new technology. Michael Cembalest at JPM Asset Management explained that “AI-related stocks have accounted for 75% of S&P 500 returns, 80% of earnings growth and 90% of capital spending growth since ChatGPT launched in November 2022.“

Even though the S&P 500 is trading at a forward P/E of 22.8 at the end of September, which is a 21st Century record, it could be argued that it’s the rest of the market that’s overvalued. The forward P/E of the top 10 stocks in the index is well below its peak in 2020. By contrast, the rest of the index, the S&P 490, is trading at a record forward P/E of 19.5.

Warning Signs

Nevertheless, the pace of AI-related expansion is starting to slow. Hyperscaler capital expenditure rates are projected to peak in 2025 and decelerate in the future.

Even more worrying is the trend of major AI players making circular deals to pay each other to fund growth, or the appearance of growth.

In addition, the pace of AI adoption is starting to slow, which casts doubt about the enormous productivity gains forecast by AI cheerleaders (see my publication AI Productivity and the Promised Land). It’s not because AI doesn’t work. It’s because companies need time to adjust their business processes in response to the introduction of a revolutionary technology, which takes time.

The Music Is Still Playing

I began this publication by rhetorically asking if we are in an AI bubble. The answer is probably. The bubble is starting to show signs of weakness. On one hand, headline M&A deals like the one concluded with AMD is supportive of further gains. On the other hand, the recent Oracle earnings report casts doubt about the profitability of AI cloud computing sustaining elevated valuations.

Technical analyst Helene Meisler summed up the situation best with her observation that the market consensus is we are in a bubble, but no one knows when it will collapse. But everyone thinks they can get out before it pops. Everyone seems to be thinking like former Citigroup CEO Chuck Prince, who famously said just before the onset of the GFC: “When the music stops, in terms of liquidity, things will be complicated. But as long as the music is playing, you’ve got to get up and dance. We’re still dancing.”

Hey Cam, Have put any work into understanding how the changed rules around retail investors, 401ks and their access to private assets will impact markets. One of the golden rules of alpha is to get in front of fund flows.

I realize you are mostly focused on public markets, but is there enough dry powder waiting in the private markets to keep the music going for a little bit. Particularly when one thinks about private debt and infrastructure funds being set up to keep investing into the build out. One (of the many) differences between now and dotcom bubble is the depreciation and utilization of the assets in question. GPUs are not lying around fallow waiting for a great expansion of demand. As long as capital flows via off-balance sheet vehicles and private dollars keep going into data centers and Jensen’s intelligence factories, model builders will build and this thing will keep going – not without its risks though.

At some point something will have to give, but I have a feeling this thing still has room to run because of this new dry powder is looking for a place to go and the balance sheets of the hyperscalers are still so big and are now just being leveraged. Open AI is hell bent on becoming a hyperscaler – otherwise it fails. In order for it to do that, it needs to invest hundreds of billions or even a trillion to get where the puck is going. Scary as it seems, that money is out there.

More and more of the GDP of the world is going onto the cloud and the current cloud is being rebuilt to accommodate this new tech stack. LLMs are the first wave and the low hanging fruit. Jensen talks about the inference market being much bigger and he might be right (or drinking his own Kool aid).

I think sensors in the physical environment, the digital infrastructure around them and the compute to support is underestimated as a catalyst and will be a multi-decade story in much the way that the Internet was a multi decade story. I am confident it will be punctuated with a crash here or there.

1 thought on “The AI Bubble Debate”

David Wilkens says:

October 12, 2025 at 6:11 am

Hey Cam, Have put any work into understanding how the changed rules around retail investors, 401ks and their access to private assets will impact markets. One of the golden rules of alpha is to get in front of fund flows.

I realize you are mostly focused on public markets, but is there enough dry powder waiting in the private markets to keep the music going for a little bit. Particularly when one thinks about private debt and infrastructure funds being set up to keep investing into the build out. One (of the many) differences between now and dotcom bubble is the depreciation and utilization of the assets in question. GPUs are not lying around fallow waiting for a great expansion of demand. As long as capital flows via off-balance sheet vehicles and private dollars keep going into data centers and Jensen’s intelligence factories, model builders will build and this thing will keep going – not without its risks though.

At some point something will have to give, but I have a feeling this thing still has room to run because of this new dry powder is looking for a place to go and the balance sheets of the hyperscalers are still so big and are now just being leveraged. Open AI is hell bent on becoming a hyperscaler – otherwise it fails. In order for it to do that, it needs to invest hundreds of billions or even a trillion to get where the puck is going. Scary as it seems, that money is out there.

More and more of the GDP of the world is going onto the cloud and the current cloud is being rebuilt to accommodate this new tech stack. LLMs are the first wave and the low hanging fruit. Jensen talks about the inference market being much bigger and he might be right (or drinking his own Kool aid).

I think sensors in the physical environment, the digital infrastructure around them and the compute to support is underestimated as a catalyst and will be a multi-decade story in much the way that the Internet was a multi decade story. I am confident it will be punctuated with a crash here or there.

Comments are closed.