Points of Return
To get John Authers’ newsletter delivered directly to your inbox, sign up here. Yes, it’s official, the US government just shut down. Artifi
View in browser
Bloomberg

To get John Authers’ newsletter delivered directly to your inbox, sign up here.

Today’s Points:

Is AI a Bubble?

It’s an impossible question to ignore, so let’s make another attempt to answer it. September saw the publication of at least two widely read analyses suggesting that what is going on in artificial intelligence is indeed a bubble: Is AI a Bubble?, published on Exponential View by Azeem Azhar, and Dotcom on Steroids by GQG Partners. 

And in line with classic bubble psychology, these serious attempts to diagnose the investment climate have appeared just as popular fears of a bubble have subsided. Adrian Cox of the Deutsche Bank Research Institute uses a battery of measures to show that the “AI Bubble” Bubble has now burst. This is how searches for the term have moved on Google Trends since the launch of ChatGPT in November 2022. There has always been more interest in the risk of a bubble than in the prospect of a boom:

Crunching through major English-language media also found a peak of interest in an AI bubble last month, followed by quick descent:

A search of the term on Bloomberg News Trends, which covers all stories from all sources that are published on the terminal, finds a series of waves of concern about bubbles, with the peak coming for a few days after the DeepSeek shock in late January:

One final measure from Deutsche, looking at posts on Reddit, where all those most fascinated by finance are to be found, shows a peak in August followed by a decline:

Cox argues that this is wholly consistent with a burst bubble lurking in the near future: “While there may be a bubble, the moment everyone spots it may be the moment it is least likely to burst.” In the run-up to the epic dot-com implosion, he points out that the Nasdaq Composite index suffered seven separate corrections of more than 10% before finally reaching its peak in 2000.

1999 and All That...

This raises the issue of the most obvious parallel for the AI excitement. A quarter-century has now passed since dot-com bombed; does this mean that its lessons have not been learned? The GQG report suggests that it does:

It may be hard for investors to face the uncomfortable reality that the trade that worked for over a decade may be over. After all, most money managers today do not carry the scars of the dot-com era. Of the approximately 1,700 active large-cap US portfolio managers, just 4% invested through that period. There is a difference between living through a downturn and merely reading about it. 

As someone who also lived through the dot-com bubble and burst, I agree with this, but with two caveats. First, I have probably been too ready to see other imminent disasters in the years since. We are all prisoners of our own experience, and while most active managers don’t have direct experience of dot-com, the commentators they read generally do. 

Second, this episode  feels different from 1999 and 2000. At that point, before de-industrialization had begun in earnest, before Vladimir Putin had arrived in Moscow, and while the Twin Towers of the World Trade Center were still standing, the US and its economy seemed invulnerable. Optimism and excitement were everywhere and frothed over. This time around, the excitement over AI feels more like a collective grasp for a lifeline. Psychologically, it’s different.

A further specific issue with benchmarking to the dot-com era is that it was by most measures the most excessive stock market speculation there has ever been. Finding that the AI bubble isn’t as bad can miss the point, as nobody should want speculation to grow anything like that extreme. Stocks fell some 70% after the top in 2000; that doesn’t mean that anyone would be happy with a fall of 60% from here.

Symptoms

Irrational exuberance, as famously diagnosed by Robert Shiller ahead of the dot-com bust, is not the same as stupidity. Bubbles are hard to diagnose, but in hindsight the biggest were in technologies that proved to be revolutionary —  canals, railroads, cars and the internet. They do, however, rely on cheap money.

Robert Kindleberger, in his classic Manias, Panics and Crashes, produced a schema followed by all the great bubbles. Here it is, as summarized by Chris Watling of Longview Economics:

i) cheap money underpins and creates the bubble;

ii) debt is taken on during the bubble buildup, which helps fuel much of the speculative price increases (e.g. buying on margin);

iii) once a bubble is formed, the asset price has a notably expensive valuation; 

iv) there’s always a convincing narrative to “explain away” the high price. 

Humans find it difficult to extrapolate trends into the future, and once easy finance is available, justified excitement about something transformative can turn into wasted money. Valuations already look overdone, as Points of Return covered yesterday, and financing is cheap, even with complaints that the Federal Reserve is too hawkish. And the narrative, we’re all aware, seems mighty convincing.

The symptoms of an imminent burst are there. GQG points out that companies previously regarded as invulnerable are suffering a steady derating even as their earnings continue to grow. They offer Adobe’s performance over the last five years as a clear example:

The money being spent on capital expenditures also suggests that many have already taken extreme risks. Azhar offers this chart of hyperscalers’ expenditures, using data from Citi:

Such a rate of increase looks unsustainable. It is the spending sparked by the bubble, rather than the share prices themselves, that looks dangerous. To quote GQG: 

Big tech CapEx as percentage of Ebitda is now running at 50%-70%, which is similar to AT&T’s 72% at the peak of the 2000 telecom bubble and Exxon’s 65% at the peak of the 2014 energy bubble. Historically, companies experiencing higher capital intensity tend to be structurally poor investments. 

In both the telecom and energy bubbles, an exciting new technology (internet for telecom, shale for energy) justified unprecedented levels of investment. Eventually, supply outstripped demand, and the companies never earned a return on their investment… However, customers benefited massively from cheap internet and energy.

Consequences

The dot-com bust happened before Google had gone public. Facebook and Twitter did not yet exist. Netscape Navigator and Microsoft’s Internet Explorer were still fighting to be the dominant browser. The money lost by the irrationally exuberant back then did not stop the internet from transforming society and the economy. And several of the leading companies from that era remain in the lead today. However, Ian Harnett of Absolute Strategy Research makes this important point:

The lead players today are still some of the best supported companies from that era — Microsoft/Apple/Oracle/Amazon. However, that didn’t stop each of those mega stocks falling -65%, -80%, -83% and -94% respectively vs their Tech bubble peaks to their troughs. Another salutary lesson is that they took 16, 5, 14 and 7 years respectively to regain those 2000 peak prices!

If you are shareholder, then, a burst bubble could be really bad news. However, the fact that this bubble, like the dot-coms, has been funded primarily by equity is good news for everyone else, as the economic impact should be reduced. Harnett suggests that the correction when it comes will be more like the fallout from the dot-coms than from the financial crisis of 2008, which was driven by defaults on debt and was far more serious for the economy. It should also mean that central banks need do less in response.

AI probably will have profound consequences on the way we all work. But Schumpeterian creative destruction being what it is, there will be some pain ahead before we all enjoy the businesses that are being built.

Powering AI Dominance 


Data centers have become critical infrastructure in the tech industry’s quest to develop the coming generation of artificial-intelligence models. No surprise, then, that companies are investing billions in acquiring the computing power they believe will drive the next significant surge in productivity. The five largest US “hyperscalers” are projected to have a combined $736 billion of capital expenditures in 2025 and 2026 alone, according to Goldman Sachs:

Source: Goldman Sachs

The bank projects data-center demand to surge from today’s 62 gigawatts, which comprises cloud workloads (58%), traditional workloads (29%), and AI workloads (13%). By 2027, AI is projected to account for 28% of the overall market, while cloud computing is expected to decline to 50%, and traditional workloads to fall to 21%.

Much of the latest data-center demand is driven by AI labs such as OpenAI, Anthropic, and xAI, which continue to see rising valuations, according to BloombergNEF. Its analysis shows that the capex of Amazon, Microsoft, Google, and Meta tripled to $217 billion in 2024 from $69.4 billion in 2019. This has mainly been financed from cash flows, but developers are starting to tap capital markets for $20 billion project finance packages. 

The optimism is reflected in the performance of exchange-traded funds that track data-center infrastructure companies — from power generators and cooling systems to servers, networking gear, and construction firms. 

There’s been similar growth in electricity consumption. All these new facilities put a strain on the grid unless power producers can match their output. Otherwise, they’re likely to drive up costs. Bank of America’s analysts, including Andrew Obin, forecast AI electricity demand growing at a 40%+ compound annual rate.

Further, there is a question mark over how this investment will be returned. User numbers continue to rise, but paid conversion rates are low, and even paid users are likely loss-making, BloombergNEF analysts argue. Whether profitability can be achieved will be the key to continued momentum in the energy and infrastructure sectors.

Rising electricity demand presents a strong case for exploring more affordable energy sources. This hasn’t happened despite the current administration’s preference for non-renewable energy. Meanwhile, Washington’s pivot to unleash America’s energy resources has yet to make a significant impact.

Gavekal Research’s Louis-Vincent Gave notes that in the face of rapidly rising electricity demand, the priority of any elected official should be to drive energy prices lower. Ultimately, cheap energy remains the most certain path for the US to assert AI dominance. One option, Gave argues, would be to push for a marked increase in oil and gas production from Venezuela or Iran — ideally both:

Despite their vast proven reserves, the nature of their regimes deters investors from committing the capital needed to optimize output, and so puts downward pressure on global energy prices. The temptation to encourage regime change must thus be strong, although it should be noted that the US-led regime change in Iraq in 2003 was associated with oil prices surging rather than collapsing.


Then there’s coal. It’s relatively easy to mine, transport, and burn, and coal plants can be built quickly at competitive costs. This explains President Donald Trump’s recent tirade at the UN, where he ridiculed climate change and touted coal as a game changer.

As data centers guzzle more electricity than ever, the price ripple is hitting households and investors alike. The pressure to cap costs isn’t just urgent; it’s unavoidable. The outcome of Washington’s pivot will now determine whether the US maintains its edge in AI or allows high energy bills to become the next bottleneck.

Richard Abbey

Survival Tips

Returning to underrated albums: Try Mogwai’s As the Love Continues, Linkin Park’s A Thousand Suns, Dog Man Star by Suede, Spirit of Eden by Talk Talk and Disintegration by The Cure. There’s more where these came from...

More From Bloomberg Opinion

  • Adrian Wooldridge: Brexit and Forget It? At Your Own Peril
  • Allison Schrager: America’s Elite Universities Have Lost Their Way
  • Paul J. Davies: JPMorgan's EA Debt Isn't the End for Private Credit

Want more Bloomberg Opinion? OPIN. Or you can subscribe to our daily newsletter.

Like Bloomberg's Points of Return? Subscribe for unlimited access to trusted, data-based journalism in 120 countries around the world and gain expert analysis from exclusive daily newsletters like Markets Daily or Odd Lots.

Like getting this newsletter? Subscribe to Bloomberg.com for unlimited access to trusted, data-driven journalism and subscriber-only insights.

Want to sponsor this newsletter? Get in touch here.

You received this message because you are subscribed to Bloomberg's Points of Return newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.
Unsubscribe
Bloomberg.com
Contact Us
Bloomberg L.P.
731 Lexington Avenue,
New York, NY 10022
Ads Powered By Liveintent Ad Choices